---
categories:
  - "[[Meetings]]"
type: []
date: 2026-07-15
org:
loc:
people:
  - "[[Darsh Shah]]"
topics: []
granola_id: 6d9f7531-851d-41be-a55b-96b1cf7970b0
source: granola
---

# Presentation Framing and Naming
## Notes

## Summary

### Presentation Framing and Naming

- Drop Merlin from the deck; it’s a codename, not a shippable product

  - Establish the Merlin-to-map connection once at the opening, then move on

  - “Asset gen” proposed as the replacement term (formerly Merlin)

  - Canvas name also unclear: sounds like a UI feature, not the monetizable model access

- PCG to be removed from AI features list: not being monetized as part of this launch

- Consistent naming critical throughout: different words for the same thing will lose the audience

- Framing for revenue model: use neutral “free tier + monetization” rather than “subscriptions”

  - Subscriptions vs. top-ups debate unresolved; Tim historically preferred top-ups for Merlin, subscriptions for UE

- 5.9 confirmed to include unreal animation framework and mobile enhancements

### MAU Numbers and Cost Modeling

- Show only one MAU figure per product: 20k effective MAU (UEFN), 600k (UE)

  - Showing both invites questions; define “effective” inline in case 800k number comes up

- EDA usage distribution (from non-prod/tool-calling data, not MCP):

  - 60% casual: ~1 conversation/month

  - 36% regular: ~1 conversation/week

  - 5% active: ~1 conversation/day

  - 1% power: ~8 conversations/day (each conversation = 8-10 back-and-forth)

- 30% UE attach rate flagged as needing rationale: licensees likely use Claude/MCP directly

- Asset gen modeled at 50% of EDA MAU for UEFN, ~30% effective MAU for UE

  - ~1.4 3D models + 2 images per generation session

  - Asset gen cost roughly 1/4 of Fab AI model (which assumed 4 outputs vs. 1 here)

- Net projected costs: ~$46k/month EDA (UEFN), ~$58k/month combined UEFN; ~$553k/month at UE launch

- Tim’s known stance: everything should be free unless it bankrupts the company

### Data Credibility and Presentation Strategy

- Tim previously rejected spreadsheet modeling: “I don’t want to see your fucking made up numbers”

  - All numbers need a sourced rationale, not just a modeled assumption

- EDA usage curve has a real basis (non-prod database); asset gen numbers currently estimated

  - Meeting with Nick Reitman tomorrow to pull asset import data from UEFN/UE content drawers

  - Greenie (UEFN) has WAU engagement buckets that could anchor power-user modeling

- Recommended structure: add a user segments slide before the cost slide

  - Show casual/medium/power tiers with actual data, then apply to cost model

  - Be explicit: label which numbers are data-backed vs. best-guess pessimistic estimates

  - Frame worst-case as “highly successful, high usage” scenario, not a failure mode

- Pilot data (internal EDA) should anchor the story; external pilot with UEFN partners as next step

  - Validate caps against actual partner usage before broad launch

### Caps, Monetization, and Abuse Risk

- Proposed UEFN 4.3 launch: free with generous caps (88 conversations, 120 assets/month)

  - These caps represent the top 1% usage tier; effectively no one would hit them

  - Worst-case if all 18k effective MAUs hit caps: $2.8M/month run rate

- Caps strategy debate:

  - Too generous: no signal, perverse incentive (Roblox devs exporting free assets)

  - Too restrictive: users frustrated, tool feels useless

  - Preferred framing: set caps where power users occasionally hit them, creating upsell signal

- EDA caps should be separate from asset gen credits: don’t penalize conversation usage

  - What counts as a “credit” is unresolved (3D jobs only? materials? Blueprint code?)

- Asset gen abuse risk flagged: without export restrictions, free assets could flow to non-UEFN creators

  - Technical mitigation: disable or restrict export for AI-generated assets

  - Community angle: if assets are private, Epic is subsidizing competitors; if shared to Fab, significant implementation overhead

- Asset gen less “mandatory” than EDA: alternatives exist (Scenario, Meshy, etc.); EDA has no real substitute inside the editor

### Trust, Safety, and Age Gating

- Age gating: strong lean toward 18+ for all AI features at launch

  - Claude is not an under-18 product; model licenses (e.g., Stability) require risk assumption for mature content

  - Gemini recently tightened restrictions; landscape is shifting

  - Tim historically fine with mature content for 18+ but opposed to children being generated

- Moderation approach: AI moderation assumed for asset gen and text

  - EDA prod safety system currently disabled in agentic EDA due to high false-positive rate

  - Classifier accuracy for nuanced content (e.g., context-dependent themes) is technically hard

  - Licensees include Epic investors making games with sensitive themes: over-moderation is a real risk

- Gwenny assigned the under-18 slide for the Tim meeting

  - Options to present: parental toggle vs. feature restrictions for under-18 users

  - Marcus should be in the room when trust and safety decisions are finalized

- Recommendation: keep 18+ simple for 4.3 launch; easier to loosen than to add complexity later

### Next Steps

- 

**Get asset gen usage data from Nick Reitman**

Meeting tomorrow: pull number of UEFN/UE users with assets in their workflow and content drawer import volume.

- 

**Pull WAU engagement bucket data from Greenie**

Use UEFN power-user hours data to anchor the usage distribution model.

- 

**Add user segments slide before cost model slide**

Show casual/medium/power tiers with real data; label which cost assumptions are data-backed vs. estimated.

- 

**Align on consistent naming across the deck**

Resolve "canvas" vs. "asset gen" and audit every slide for terminology drift before the Tim meeting.

- 

**Ensure Gwenny's under-18 slide is ready for the Tim meeting**

Slide should present a menu of age-gating options for Tim to choose from.

Chat with meeting transcript: [https://notes.granola.ai/t/45f69f5f-b4bd-4f02-8975-4f22f65e2656](https://notes.granola.ai/t/45f69f5f-b4bd-4f02-8975-4f22f65e2656)

## Transcript

**system**: Can we drop Merlin from the like I feel like if you want the focus to be on 43, Merlin feels like a big distraction. I 100% agree with that.

**system**: Yeah, I think honestly I agree that considering the two partners who are also top investor of epic and the costs are going to be very marginal, there's no point raising bringing that to anybody.

**system**: And the other column tooths of the table. Yeah, that one to me is five nine.

**system**: Experimental.

**system**: No, we have to keep this because this is this is effectively EDA in release.

**system**: Right, but we should call Berlin experimental then. That's like way undersells it.

**system**: Price from the oven. I just had my sync with Marcus and we have a 5.9 coming where there's going to be unreal animation framework. So it's going to be mobile enhancements. There's going to be great. This makes sense to me. So it was written in the sky. Yeah.

**system**: And I've told him again that look.

**system**: There's no merlin we are shipping canvas, EDA and cementing and whatever is needed for merlin to operate.

**system**: But what matters are those components and that's what we have to highlight.

**system**: Yeah, we should come back. If that's plan of record, the branches are a mess right now.

**system**: It's very hard to do that.

**system**: Yep. Let's come back to that. But anyway fundamentally, Mark, that took a lot right.

**system**: Calling in five nine and we're not going to get a whipping. The other thing here I want you to check is the is the effective maus here for ue, I think 600k is right, but I want to make sure that that's that's true. What is what is affected? I don't know what that means. That's like reaching some hour threshold inside the editor. So for example in uefn that's greater than 60 hours in a month in uefn. So we have a top line mau of 30, but of the people who are like actively engaged. It's 20,000.

**system**: And I think does it mean anything to show both?

**system**: Like what's the inference I should draw from.

**system**: That? The only reason I wanted to show both is because I'm afraid that people have heard different numbers at different times and I didn't want to present one number and then somebody like, no, no, I heard it was 829,000 users. And then we say, well, that's the ma versus the effective me.

**system**: U. I'm fine.

**system**: Not show both. I just didn't want to introduce any more. We must have covered this in other meetings before. What do we typically do?

**system**: Yeah, I'm not sure. So yeah, I think we need to my question mark is which number are you using for computing or making the simulation of the cost, et cetera? Effective.

**system**: So just said we have 20k ma you and 600k. Like don't show both because that is a door for team to ask questions.

**microphone**: Yes.

**system**: And I don't even want to get there with him.

**system**: We are considering 20k mau and 600k mau and that's it.

**microphone**: I would just say maybe we can not the mauber mark and just like define effective there in case somebody brings up that like, you know, they've known 800k to be the number.

**microphone**: Something like that.

**system**: And then the last line here is just to not bury the lead is just to say here's what we're proposing for the revenue model.

**system**: That we're going to go over in the next couple slides.

**system**: Again.

**system**: I've had it backfire showing it earlier. I've had it backfired not showing it earlier. And so I'm just open to suggestion here.

**system**: I think.

**system**: The free with generous caps has to mention like rate limiting or something like that.

**system**: Because to me it's it's really that's what we're saying is that it's like it's a generous rate limit or something like that.

**system**: Well, I guess that's a question cap per day.

**system**: Well let's go into the numbers and then because I think we're going to end up maybe deciding some things here that we want to go back and feed into this slide. The idea is sort of a cap per day, but that doesn't actually really work all that well. So it might be like a cap per day per month.

**system**: Or something with an additional cap per month.

**system**: So let's maybe come back to this one.

**system**: I was just going to question subscriptions as now because it's I always it felt like top ops was like the framing that Tim liked.

**system**: And that's different than the subscription model, right? Yeah.

**system**: So I remember this is I what I remember is when Merlin was going to launch that we were going to do Merlin with top ups.

**system**: We were not going to do but but I always thought the plan was subscriptions because it ties in with UE subscriptions and the UE teams.

**system**: When we actually got to launching inside UE.

**system**: I may not have been a part of those conversations or maybe I'm misremembering.

**system**: I mean I think Eugene was the one driving those conversations so we may have lost some of that context. I wasn't in those meetings either.

**system**: I thought I thought the tacos was really just a Merlin thing to try and get Merlin out the door and be like, hey look, it's just going to be top ups. You don't need to make anything more complex and then we do subscriptions later. But I think I mean for this slide I liked putting this up front. I would recommend it's just be more neutral like free tier plus monetization. You know.

**system**: In part I think there's like another rabbit hole here which is that there's a lot of press coverage right now about how like anthropic and open air losing their shirt on subscriptions or they would be if people used it as much as they actually could use it.

**system**: And you know and then they're oh but there it's a lost leader they're training on your data and it just feels like a whole.

**system**: Thing that.

**system**: I don't feel like it's un like none of that's going to be true for you know what I mean it just feels like.

**system**: I mean we're going to stop anyway in a slide or two so but but a fair point is fine. We should keep in mind that in every conversation at least I've been in and anyone feel free to chime in Tim will always Tim comes from the standpoint of everything should be free.

**system**: Except when we have to like bankrupt the company and then it can't be anymore. So he's not going to give a shit if we're losing some of money as long as it's not overly egregious. And even then in a meeting he's not really going to care. It's that's always us or not us specifically but the people who are presenting it who have to come back to him and say no you're going to bankrupt the company if you do this.

**system**: He always comes from the other way.

**system**: One additional comment as well when it comes to features.

**system**: I would remove PCG.

**system**: I don't think PCG is an AI feature per se. I mean for sure PCG is coming to UEFN but it's very independent from what we are monetizing. Otherwise we're going to have to start to list everything supported by tool registry and.

**system**: That's.

**system**: Within it was added to a launch scope document. So I thought we wanted to care it is part of the launch.

**system**: But it's I don't think we are monetizing PCG per se. It's what we are focusing here is really the component. And by the way as we go through the deck and we iterate we're going to have to.

**system**: Pay attention to make sure that the wording we are using is consistent everywhere.

**system**: Because otherwise we're going to like if we call different things different words we're going to lose team and so it's going to be very important that we start to stick to the same naming.

**system**: On one side and not the other.

**system**: Because it's already in UE 5 no but it would this would be the managed service semantic search. Yeah it has to be on both side because with 591 the semantics search as a managed service as well.

**system**: To your point about using the same name. I was unsure whether or not campus had been a name that we wanted to keep going with or so I was saying ask at gen for clarity but I'm happy to.

**system**: Refer to his canvas if that's the decision. Yeah it's also unclear that like canvas sounds like the editor UI feature.

**system**: And the thing that we're charging for is like the models that might be nodes in the canvas.

**system**: Let's circle back on it just we probably we probably have some edits to make here. Yeah we might have I think you know it's a good.

**system**: One I think was referring to content generation or asset generation.

**system**: And maybe that's what we refer to because that's really the capabilities. Canvas is just a front end. And in fact you could use canvas maybe even without AI again will not monetizing canvas but we are monetizing is the access to the model.

**system**: So we have to change that maybe it's asset gen maybe is enough.

**system**: But.

**system**: Do we want for 5.9 to Merlin as one of the bullet points?

**system**: I feel like Merlin is like a name for an initiative.

**system**: Like I don't see how it's done. What are we monetizing in Merlin beyond EDA and SH and semantic search? Is there anything?

**system**: I mean we might have diffusion model that you don't get with 43.

**system**: But that's just the point I would argue why not giving those diffusion model. I mean it's at this point it's.

**system**: Because if I if I talk to the Merlin guys I said you can do a lot of assets. I mean other stuff than linear content with merlin. So at this point to me it's the same it's asset gen. I mean it's.

**system**: I mean I said in general whatever you think makes more sense.

**system**: But I'm just trying to like we're presenting at 10 right and this particular slide is AI release timing not necessarily I mean it's in the monetization chunk. And so we could say it's like scoped that particular thing. But the nouns that he knows right are he should know Merlin at this point and he'll probably know asset gen because we talked about it previously with a bunch of other stuff.

**system**: I don't know if it's valuable out of here or not. I just asked the question because I think that when we set the context at the opening of the meeting and we present the map we should call out that pretty much what we call marlin is that piece is on the map and so that he understand and then we don't really refer to marine anymore.

**system**: Okay. We're going to have to move away from merlin because merlin, I mean it's a great codename for us.

**system**: But the public might never see that name.

**system**: So when said in the visualization we're going to just say like acid gen with an asterisk formerly known as merlin.

**system**: No to me.

**system**: In the diagram I sent merlin was kind of semantic plus eda plus agent plus.

**system**: Asset gen and those three shape kind of the base pillar of merlin.

**system**: Yeah I feel like merlin is like it's just a it's like a name of a prototype right or it's the name of a special project.

**system**: And that we're not going to ship that special project and so I don't think the name Merlin should really yeah I agree it's like it was just like a dotted line around some boxes right it was a prototype for canvas but that's nonetheless. I mean I don't want to hijack the entire meeting sorry mark but nonetheless during the mapping opening of the meeting we're just going to have to make sure we make that connection between what is merlin and how does it refer to the map so that you understand and then we can move on otherwise it's going to be constantly loss on that.

**system**: Yeah, we haven't even gone to the objectionable part of the presentation.

**system**: Let's go to the next one yeah go to the next one all right so here's what we start putting in some like projected costs again this may not be the best way to present it but on the left hand side we have EDA on the right hand side we have asset gen and again I'll take the notes from before of removing these extra MAU numbers but effectively you know the 20,000 here and the 600,000 here of the MAU numbers were basing this off of.

**system**: What we're trying to do here is sort of give like a what would be like a really active and engaged user base here so we're saying like what if 90% of our like our UEFN users were actually in EDA doing things?

**system**: And at that point we get 18,000 maus using EDA.

**system**: Now they're not all going to use it the same way there's going to be some distribution curve about how they use it here.

**system**: And so what we've been sort of going back and forth here a little bit on is that you know roughly 60% are doing about a conversation a month.

**system**: We have another 36% that's doing like a conversation a week.

**system**: 5% are doing a conversation a day.

**system**: And about 1% are doing like eight conversations a day. Those are the heavy users and a conversation is like eight back and forth eight to 10 back and forth right so I'm actually building something I'm doing something it's not a one off fire and forget it's a it's a back and forth here for a conversation.

**system**: Like going back to the top you have 30% on UE.

**system**: Why is that so low?

**system**: I actually think that's their pre I mean we should say check it but like license like lots of licensees are going to use you know claude or mcp or their own thing right so I would expect a lower percentage of UE users to be using EDA because you've given them a really really good option not to.

**system**: So we should yeah we should put some I don't know where the number where 30 came from but I feel like there should be some diligence and rationale behind where exactly 30 came from so I think the question is going to get asked.

**system**: Yeah that's fair question I'm supposed that you have a spreadsheet modeling all of that behind right yes okay is there a word where we can present.

**system**: A bracket maybe on a low usage high just show a kind of some boundaries here that is not just like a middle but maybe okay in one can it could cost like that but if people were let's say if it's the success was much higher with a free key where that's what we're talking about just so that they understand the the spectrum of the cost because it's when I look at the it's not really the 30% that I'm challenging because.

**system**: When you I mean what you would have to do here is really look at the.

**system**: Howly rate of usage per day and when we look at the data sometimes people what we consider MAU are people logging in like once there's a lot of nuance on the MAU but it's more on EDA usage like eight conversations a day.

**system**: 1% that seems like nothing ate back and forth for me.

**system**: That would be effectively eight conversations at eight to 10 back and forth so 80 back and forth oh sorry I thought it was eight back and forth yeah that's so I think it would really help here this touches on subs I think it'd actually be nice to have a slide before this where you talk about like like we have these user segments right you know where we basically have a sort of like low medium high power and actually just unpack that and be like hey, you know we did we did some data mining and this is what we see right now right and then that gives us a chance to sort of just say yeah and like at the high end that's like a conversation an hour with multiple back and forth right we could do something similar for asset gen but I think kind of framing that first then you would go to this slide and then that's where you now you're saying like okay well if you apply you know this model.

**system**: You know like you know we have this breakdown of users you know like now you know what a casual user is if we assume 60% of our users are casual blah blah blah blah blah and then you could also to set up a follow-up slide be like but maybe we're really good at our job and eda is super useful and you can imagine 10xing you know or every each of these users you know uses it 10x more than we're modeling and that would obviously tedex the cost you know something like that.

**system**: I will say my honest hesitation on that is that I did a presentation at Tim a while back on search with a spreadsheet and he said I believe the exact quote was I don't want to see your fucking made up numbers in a spreadsheet.

**system**: So I'm a little cautious about kind of like hey this is the model that we did and stuff like that because he doesn't necessarily so that's why I also think like there needs to be any number he puts on here whether 90% 30% 58% 36% there needs to be a how the fuck did you get that number?

**system**: And we need to somehow convey that and that's why these numbers that you're highlighting these are not made up right these are come from existing use about that when we get to these numbers on asset gen these are made up numbers or again well we have why we have gen media data right we have workflow builder like we have we have an analog of this already how are users that like I don't know that that's necessarily what we.

**system**: Want to mode there's a distribution there as well but at least it's something right if we have any data there at all.

**system**: Yeah also I suspect I mean again that's my perception but I suspect that people like if you look at the population of subject matter experts on UEA is much more versatile gen assets is not everybody needs to generate 2D mesh and image as well like there's a lot of people who are not going to touch chain assets as part of our user versus EDA that is everywhere in a way yeah but I agree like it's very hard to go into any of these meetings and say yeah we've made up some numbers and now we're modeling based on these made up numbers like any data we have I think is is going to be better than that.

**microphone**: So.

**microphone**: I just wanted to like add one point here. Like as I kind of put these numbers together, the percentage of users adopting EDA and asset gen that is one place where we don't really have a very strong like scientific basis of getting at like an estimate.

**microphone**: So would love like any ideas there from the group. Like if there's a way we could like, you know, maybe just estimate that on EDA usage like Nathan and Mark said like those are like from the non prod database, which is basically the core basically the tool calling version that is currently being used.

**microphone**: This is the usage curve we are seeing there barring the last one which is the 1% that's just a tail that we added to kind of account for the fact that like we'll see a lot more usage when we release the full tool.

**microphone**: On asset gen these are like we just funded these estimates right now but I'm meeting with Nick Reitman tomorrow to figure out a way to get the number of users who today within UEF and UE have assets as like a part of their workflow maybe the number of assets they are importing into the content drawer and like bringing into the editor.

**microphone**: Based off of that we'll come up with a more like polished estimate there for the asset gen usage.

**microphone**: So we'll have like a basis for both of those.

**microphone**: But for the for the piece where we are like estimating the number of AI attached users, that's that's one place that.

**microphone**: Doesn't really have a very like strong.

**microphone**: Basis as of today.

**system**: If I why greenie on the UEFN side she maintains that they have engagement bucket and you can see the.

**system**: Like wau in this case and the number of hours they spend.

**system**: So that's also an interesting data like we don't have that many power user when you look at UEFM.

**system**: Yeah so that that's another interesting data point if you want to model a bit what you think is going to be the usage again you we can make that data is available so it's up to you just just to be clear when we're saying EDA usage statistics for cor that doesn't include mcp.

**system**: Right Correct.

**system**: No but we do have those numbers if we want them I'm just saying.

**system**: When that question gets asked I want to make sure that we can answer that.

**system**: For the deck market that's what I get I think given how heartle given at least on the eda side you know I think since we have a lot of sight to asset gen like the sort of the segments we actually have pretty strong evidence for I would pull that out talk about the segments you could also the cost is another thing we have pretty strong evidence for right and so just talking about like introducing the segments introducing the notion of a conversation also here you have you have cost per query but everywhere else it's it's in terms of conversation so that it was a late change here and I want to sort of homogenize this so we have questions per month and then tokens per question and whatever but I think it would do it in conversation context is probably where that's a better metric.

**system**: But then and to me I think like for you know things like the attach rate like is it 90% of fortnite users that's just inevitably like a finger in the air and to my mind I wouldn't hide that I would just fly right into the teeth of it and be like you know God only knows this is our best guess here's why.

**system**: You know and assuming it's probably better to overestimate than underestimate when we're talking about costs underestimating revenue but overestimating costs.

**system**: So you took notes marked that maybe we have to split that into multiple slides just as we discussed you said the table and I think be very explicit around like we have solid data on certain item and for others.

**system**: Like best guess and we are pessimistic in a sense that well it's going to be highly successful with a lot of usage.

**system**: And that's that's a starting point.

**system**: Because I'm correct me here but I'm assuming it's pretty linear in a sense that if we double usage the monthly cost with double pretty much is that correct?

**system**: Yeah basically okay so then it's it's kind of easy to to extrapolate unless you're shifting within the buckets right you're moving from yeah, but that's one is a part who seems to know that that one is to be the part where you have some data.

**system**: Okay.

**system**: So the net result on eda here is about 46,000 a month.

**system**: Given these numbers to run EDA in uefn in ue obviously it's much higher with much higher ma use.

**system**: On the asset gen side we basically took 50% of people who were using EDA would also generate assets right so some smaller number there which is 45% of the total MAU and I need to maybe make this a little bit clearer because it is not exactly that way on the on the UE side we did like 30% I think of the of the effective MAUs which ends up being about 12% of users of total maus who were doing asset gen.

**system**: Given the distribution the other thing to note in here is like what does a generation mean and and so we were sort of modeling this out is like well it's two images it's generating image and you edit an image and then you generate a 3D model and some of that time you're going to do something else at 3D model whether you're going to retexture it or decimate it or skin it or something like that so that's why we get like this 1.4 3D models and two images that's that's sort of what these numbers come from.

**system**: So one of the sorry one thing I just realized that I didn't catch before was that you know you talked in the previous slide that there's caps but you don't actually say what those like is is it implicit that the cap is for the one like based on the EDA does everybody get.

**system**: Enough tokens to be in the 1% so we cover the caps a little bit later here is just a way to sort of limit runaway costs and we that's where we need to determine what those what those caps are these are not caps these were these are sort of like.

**system**: Uncapped what our cost we expect to be.

**system**: Like given this distribution this is what we think would happen but it doesn't necessarily mean we're stopping at 46,000 a year that's not the that's well there's no P l concept here it's not about like how much we make or it's just like this is a spend based on what we think is going to be the usage yeah did I answer your question Nathan okay on asset gen how are those numbers in terms of usage etc further away from what you had modeled for fab ai.

**microphone**: This is a pretty much in line with what we had for fab ai as far as I remember the numbers of the top of my head.

**microphone**: Cell.

**microphone**: Maybe like there's some variation just because fabai assumed like four outputs.

**microphone**: At a time and here we are assuming like just one output. But like the mechanics of it are pretty much the same as Fab AI.

**microphone**: But I'll also just look at like what the top line cost was versus what we have here just to kind of make sure we are not like way off.

**system**: And in fab ai we were also assuming that we were looking at ue maus right not uefn web based and that was kind of the closest yeah.

**system**: Okay I'm surprised because when I look at the number I was expecting for some reason asset gen to be much more expensive than ED at the end of the day.

**system**: I think video is the most expensive piece of it yeah but I don't think we have any video here correct I mean there's there's a pretty big difference in terms of usage on these charts right two generations per month.

**system**: Versus 58% well it's a 36% of conversation per week.

**microphone**: Yes.

**system**: That's what's different than I thought.

**microphone**: I just pulled up the fab ai model real quick and we were at approximately 500k a month on the revenue side.

**microphone**: And kind of similar range on the cost side including storage and moderation.

**microphone**: This is roughly one fourth of that so it kind of checks out it's just because we assume one.

**microphone**: Generation per.

**microphone**: Like step.

**microphone**: As far as like the number of users goes again like we are pretty much.

**microphone**: At around.

**microphone**: This. So we had a higher assumption there but we had like a broader top of the funnel that was like sitting in the low engagement cohort.

**microphone**: S.

**microphone**: So it's all kind of balancing out and the cost here is basically just a one fourth of what we had for fab ai.

**system**: It is a good point though we don't include moderation costs in this.

**system**: At all.

**system**: And maybe maybe we should.

**system**: We sorry.

**system**: Do we need to have moderation here because this is not about.

**system**: Publishing content and Ireland like other game developer if I want to do an adult game we should not restrict that I mean so yeah this is this is like trust and safety disagrees and there's actually a conversation that I tried to tee up yesterday it's it's separate from this but I think there's a lot of desire to like pre moderate and I think that's let's just be clear Marcus has been very adamant that we shouldn't do that and he should be in the room when those conversations happen so we will come up I think as part of this meeting with Tammy I think they'll be there I do believe that there's a little bit of difference though between UEF and UE and we probably need to at least acknowledge the difference between the two.

**system**: In at least in all the things that we've dealt with from the fab side and UFN we've dealt with it as closer to fortnite then obviously I mean to me the difference is that we we moderate publishing in UEF and we don't moderate publishing with ue but I don't think we should moderate creation for either of them.

**system**: 18 users sure a few bright lines for obviously illegal stuff but like.

**system**: The more nuanced you know like oh that's a little off cut we don't allow potty humor you know but I mean I think that when it comes to age gating everything AI will have to be.

**system**: Above 18 because claude is not an under 18 product anyway.

**system**: So I don't know that well I don't want to put words in his mouth we've gotten strong pushback from Tim on that on before on what sorry does it.

**system**: For I mean roblox allows you 18 users to do it like it feels like.

**system**: He's going to want it okay so well but but there's a difference where roblox has their own models and we don't we cannot violate licenses of any of the models that we're using.

**system**: There are a few models that we use that allow it I think it was brought up in the meeting with trust and safety that Gemini has recently reduced it like stability allows you to do it as long as you're taking the risk for it.

**system**: Yeah so I mean I would say sorry to Nathan's point like whatever whatever the restrictions that we have to abide by are legally we should abide by those right.

**system**: Have you mark have you talked to trust and safety around like age gating and all those type of things related to yeah we talked about it on on Tuesday with them they're they're gonna I actually assigned the the the u18 slide to Gwenny who's gonna take that part on some of the stuff that they had was just like they'd come up with kind of a range of options and again like in the in the fab UI if I case it was tim had said like I don't want images of children being generated but he was fine with with mature content being generated again it was all going to be over 18 I don't think there's necessarily a big push to make this like playground safe I just think that there's probably some lines here that we need to define that are that it will make it clear but I'm not getting the sense that they want to make this like a Disney type experience but but to the point the question came up as to whether or not we actually do like parental controls like a parent needs to toggle it on for a child or whether we just limit some things if you're under 18 you can't upload images or you can't do certain things it sounds like they kind of want to put that menu in front of Tim and have him pick kind of where he wants to go with it.

**system**: As long as they are working on the slide with you and ready to present that as part of the meeting I'm fine but because I'm far from being an expert on trust and safety and all those aspects so my personal vote would be that for a first launch in 43 we keep it above 18 and we make that as simple as possible it's easy to add more but just tacking much more trust and safety stuff initial launch is just more and more work and scrutiny.

**system**: And I'm assuming for the moderation piece.

**system**: That's AI moderation or is that human moderation?

**system**: So it's AI moderation on the asset generation I assume will be true for the for the text if there was any in the text as well I think yeah that's right I think the trick is that we sorry there's like the EDA prod actually has a safety system but it's actually the first thing that we turned off with the agentic EDA because it shoots he has a lot of false positives and I think the challenge is basically getting your false positive false negative rate right once you introduce that kind of like you know it's one thing it's easy to be like you know if you're under 18 you can't upload pictures that's just like a bit of logic but where you're like running a classifier you know on every prompt or every image and your criteria is sort of like you know.

**system**: Are then is it about Nazis and are the Nazis the good guys or the bad guys if we shade into that that's where it's going to be very hard to get the.

**system**: Like just technically difficult to get the classifier right and we have licensees including our own investor making game with netizen on top of that so that's a type of example for the creation site starting to put moderation is very tricky.

**system**: I think.

**system**: Anyway we'll see how it goes.

**system**: But but again Nathan brought up the point in a very good way with trust and safety that that that we do we do not want to be overly moderating this stuff and it seemed like Sarah was aligned with that we just need to make sure that we keep the.

**system**: The foot on the gas there Michael to your point having Marcus in the room when that gets discussed probably will drive it to a conclusion that we would want.

**system**: Okay so the the sort of thing here to remember or to look at here is you know starting off the first month UEFN is about 58,000 a month for asset gen and EDA and then that sort of be like yeah as user grows and that we don't exactly know what that growth rate is it currently UFN is not growing so but let's assume you know some growth there that number goes up but then when we hit the the launch of of this stuff in UE that's where we see this sort of 10x number go up right we see a 553,000 a month so that leads to kind of the proposal that we want to have here which is that we release UEFN 43 with free with very generous caps and so this these caps here are sort of you know 88 conversations and 120 assets a month that's like that top 1% tier right that's that's where you get to it and maybe we want to try and get that above it.

**system**: If everybody did that if all of our effective maus all of our 90% of our effective ma use were to do this and and just blow it out.

**system**: That's a 2.8 million per month run rate in uefn.

**system**: So 18,000 users doing the 1% here that's that's what we'd be looking at that's not likely yeah but what I mean I'm sorry what I don't get it's maybe just a format is.

**system**: Where you place that limit that gap what does that represent as part of the projected population that's why I have a hard time understanding like with 80 conversation 120 asset amounts.

**system**: What is that like it's 90% of our user base.

**system**: That would be.

**system**: That maybe miss misunderstood the question well I think yeah this is where it'd be really nice to say like if we have these segments right from casual to power user are we saying our cap is that everybody is a power user are we saying our cap is that like everybody is a high user but I think that's the light because right now you have to kind of connect those dots and do the math in your head which is a little bit hard.

**system**: Er every user being in the top 1% of what we're projecting our users are and even being generous above that.

**system**: Right 120 generations is double what we assume the 10% is.

**system**: So the most generous it would be the highest usage beyond the highest usage of what we're estimating.

**system**: Where does the number 88 come from?

**system**: Where do I say that?

**system**: Yeah next slide sorry yeah okay totally fair point that is backing into eight conversations.

**system**: A day no doubt.

**system**: Yeah that'd be like 160 you know a month yeah.

**system**: I don't I don't have the I don't have the right numbers called out here because I do it in questions per month and shit like that so here yes I need to correlate that use the same terminology everywhere yeah yeah yeah yeah so it was coming together late so which is which is why there's this disparity here but yes that is effectively like top max so the spirit though is that like everyone like that like if everyone is at the 1% no one would hit the caps.

**system**: Like we're capping nobody is never going to pay on ufn that's what you're saying no but it's like we we're sitting the cap at the absolute high end of our predicted.

**system**: You know usage yes we're setting like whatever we think a power user will do we're doing that and in fact you'd have to introduce a power user tier for asset gen.

**system**: For that to be true right there's actually double your 10 right now.

**system**: And and I don't even know that that's the right framing for this here like I like there's nothing magical about showing it as 2.8 million there's like I mean this is where I like some feedback about like how how would how should we present this I mean you might want to call back to the number you had on the previous slide I mean force of it which is whatever 50 something thousand a month 58k a month and say hey look with this cap this is our expected spend and if everything goes really bad or well depending on how you're defining it the worst case would look like this yeah I think because right I definitely had like whiplash where it's like 58k that's not too bad and they like or maybe 2.8 million.

**system**: Is it worth calling out doing the experiments with the like tongue and some of the other partners as a validation of some of these numbers.

**system**: Just as like a gate thing here to be like we're.

**system**: And just sanity checking this we're going to throw it at some what we think our heavy users and measure what we think the actual numbers or see what the actual numbers are and certainly if they're going to be like a launch partner or have some write up like we want them to be able to have created whatever they created within the caps in which we're going to put in. Yes we need to make sure it all fits.

**system**: Yeah so I agree with Shah I think it's worth having that as a bullet somehow.

**system**: Yeah I would actually try to like in a sense make that I think it's a really great it's just a really important having that be actually the core story of like hey you know we have at least on the eda you can say we had this pilot internally and here's the data we collected right you know the next step is we're going to like have a pilot externally with UEFA you know assuming that you know.

**system**: That like here's the numbers assuming that that's true here's how the numbers cash out.

**system**: But I think this sort of like try to have an intermediate step between our internal pilot and just everybody gets it.

**system**: Yeah and that'll help with the sort of made up number side of this right which is like yeah we're sort of making up some numbers here but we're going to know what these numbers are before we get any further. I think I'm just repeating what you all are saying.

**system**: I also want to concerns I have here is like.

**system**: The optimal.

**system**: Like.

**system**: If we actually do this and anyone on the internet finds out the smartest thing a roblox developer can do is like sign up for one of our accounts and figure out how to export assets out of UEFN.

**system**: Like I think you have free like the EDA I don't think know that the abuse factor is very very small.

**system**: The asset gen like unless we can unless we have a technical way to lock down assets that are in the UEFN project so that you can't export them. This is like a massive giveaway to all non UEFN creators.

**system**: Though. What?

**system**: How like other than some terms and conditions.

**system**: In usage, how would you prevent that?

**system**: So I think there's two things. First off we could certainly disable export you know from the editor for assets you know with a certain type of people around things we could at least make it inconvenient to export assets that had gone through this pipeline.

**system**: But also I think even to set the caps at like higher than the power user to me.

**system**: Like I feel like that's a little bit too like we don't get much signal that way.

**system**: Right like I think we'd want to set the caps to be generous but not excessive.

**system**: Because you in a sense that you want people you want to know like is everyone hitting the cap is like pros are hitting the cap but not other people but trying to get some actual like like I think if you're too if you're too much of a scrooge then everyone's pissed off that your tool basically does nothing right but I think if we're too generous not only are we like creating these perverse incentives you know but we're also missing an opportunity for people to say no I want more here take my money I would pay you for it.

**system**: Right.

**system**: Yeah, I don't know if that means that we have to launch though like at least the way the conversation went in fab ai and again I'm not necessarily I don't I think.

**system**: Conversations have evolved from there.

**system**: But once we introduce the cap and then was you know true about people like might would hit the cap it was well then you need to have some way for people to actually like get past that cap like you can't just frustrate a bunch of users who are like I got to wait a day to go do something like that's not real that's not a reason that's not a real solution that's why I think our plan should be exactly that like I mean unless this is a charity exercise unless we're just like epic wants to spend $5 million a month increasing the amount of like 3D assets in the world then I yeah I think like we should give stuff away for free and be ready to accept money when people want more.

**system**: You know.

**system**: I mean that's a that's a hard one.

**system**: I think it's fine to consider lower caps I think trying to suggest that we shouldn't export them out is not going to go over well.

**system**: I think as well that to me when we look at the cap.

**system**: The EDA conversation side is.

**system**: Is really mandatory and helpful to make an experience.

**system**: To operate the system etc because if you don't know you just don't know and there's no real alternative I mean unless you you go deep and you learn and you spend time asset to me is like we are providing insane amount of free asset like you have all the quick sell library the UEF and the fortnite content you.

**system**: Have hundreds of thousands of free assets on on on on fab we have four million assets on you on sketchfab.

**system**: I don't think we have to be super generous with asset per se I think.

**system**: You can go to scenario.com and and triple and meshy and you get free asset also everywhere I mean if you really but my point is that asset is is a nice one it's not as mandatory as conversation where I would be frustrated if EDA was telling me were you done you cannot talk to me anymore.

**system**: I mean that's my perception I could be wrong.

**system**: But no that really I think another way to flip it would be to say like yeah we can give away a lot of assets if our goal was to have like even more like if every freely generated asset was available to all UEFN creators right which is kind of like what the fact which means we're doing on fabai yeah right then then there's like okay well sure we just have to we're just deciding how much we want to spend to increase the number of assets in our own asset store but if these are like private assets you know our assets only for my game then I think what you're saying so yeah we're like spending a lot of money to give away something that's not really very differentiated do I mean I would the only the thing I could don't think about it from 10 perspective right I feel like Tim is going to want to give as much free as possible.

**system**: Throughout all of this at least on the uefn side because the key part of that we want for ufn is growth of users and people generating things and obstacles to that growth are potentially problematic if the cost like if cost is one thing but we haven't shown any cost here that would like be problematic for us as epic right now.

**system**: Justification I could see is if we were worried about our fab community partners.

**system**: And feeling like.

**system**: We've marginalized all the people pushing stuff into fab by making asset gen so cheap.

**system**: That in uefn that there there's no market for fab anymore because you just go live with with content asset gen right that's to me is the only community related thing that might resonate to 10 on uef.

**system**: N.

**system**: He but not really care about that on the fab side yeah it was exactly what I was going to say was not the fact that the fab community was going to be in an uphold.

**system**: And I know Dan doesn't give a damn about that either.

**system**: Okay but that's right.

**system**: Yeah this is just for like I to me if we're going to be very generous on the asset side it would be great if those are like assets for everyone you know one of these I have a 12 year old son who's doing a lot you know doing a lot of roblox work and I think one of the things that's amazing is just the amount of free assets right that just like leave it a crappy version of anything you could ever want is there for him for free right and so if our like if our ask if we're giving away kind of asset gen credits but the net result is that like all creators have like a much richer library to pull from.

**system**: Then that feels like some of that you know benefit accrues back to us versus just being like oh no we give these credits to you, you get the meshes nobody else gets them but also you could take them to roblox I mean we got it it got weird when we were trying to do that because the next question that came was like well what if you're a professional you don't want your assets going to everybody you want to try and do this for real professional work then you all of a sudden you can't do it maybe maybe it's just like hey this isn't the tool for professionals but that's not exactly what we're trying to tell but to me again I feel like we're bending over backwards to not charge people money for something that isn't free and if we just said yeah you can buy credits and if you spend if you're spending credits that you bought yeah it's yours.

**system**: You know.

**system**: Yeah I mean that was the fat I mean in the case of gen asset in engine making.

**system**: Those free asset available to the entire community would mean that they would have to be uploaded to fab meaning that they would have a product description I mean.

**system**: There's a long list of things that we would have them to implement and for the end user to text and comment on like put a name for the asset and that's something that you have to do for a product page for each asset that seems like a big effort I mean don't you think mark.

**system**: I mean we had it kind of automated from a I perspective doing it to sketchfab like generate the name it would generate the description it pull out the metadata stuff like that I think.

**system**: You're right that we have all the semantic search available to text to describe.

**system**: We were pushing that stuff to sketch fab not fab and not that that makes a big difference but it's a little bit of a different community there which which you know that that will no longer be an option. I'm going to be quite frank on the amount of free assets I really don't have a strong opinion or even a good understanding of.

**system**: Because 120 assets seems like a lot per month but you just experiment on an asset you do 20 30 iteration just to find what you want and you might have blow up most of your budget already so it sounds like a lot but it's not 120 finished asset it's.

**system**: Just iteration but one thing that this does seem to be coming out of this is that we might want to consider splitting off like a credit should at least in the UFN beginning launch phase shouldn't count against you on EDA and then be like whatever fungible across to the asset gen right like we don't want to stop you don't you don't want to generate 120 assets and then go to EDA and be like you can't do anything like we want your EDA usage to be uncapped the cap that was kind of the interesting thing is like well what does a credit mean then right is credit only for assets generation asset jobs in the sense it's 3D but like what if I generate a material what if I generate first code with a five gener.

**microphone**: You're.

**microphone**: Five hours.

**microphone**: Ago.

**microphone**: Yeah so we also take like shells.

**microphone**: Even with g9.

**microphone**: Yeah she have quad study.

**microphone**: And yeah.

**microphone**: It's for basically under the bed.

**microphone**: Under the couch.

**microphone**: And sometimes just go and like hide under this chair.

**microphone**: Yes so this is generally where she like hands out in the day but like she's been like lighting.

**microphone**: So she's going to join me kind of.

**microphone**: Hopefully when you usually find yeah she should be she should be okay maybe since she likes to just eat but yeah so he was a little food and water I just refilled cup order okay so it should like I mean well I'll probably let you know because that would be weird at the end.

**microphone**: You give to a certain amount or do you just keep it pretty cool so this I just keep it like pretty full she kind of just.

**microphone**: She's been like I've been having a tough time getting her to have like red food.

**microphone**: But like the foster actually is just delivered some like food to me which they said she might like so it's just like a lot of experimenting with that food that needs to be done so it would be nice like if you leave wet food for her one night.

**microphone**: If next my next day in front of me because she might yeah exactly just deliver them yeah like.

**microphone**: Yeah.

**microphone**: Instructions.

**microphone**: My notes real quick.

**microphone**: Yeah like that's just one thing but otherwise this is like she's te to have given whenever she said she's like good about it where do you keep the dry food for refill if she runs out so I keep it inside if you're what I'll you know just leave it out I have a box out there okay.

**microphone**: Actually is already there so I'm going to tell this show that to you.

**microphone**: Yeah like the ladle is pretty full as in like the sanity yeah yeah we should be okay against like yeah just cleaning that maybe once in the evening whenever you're visiting up here I'll just yeah exactly so today I'm just going to clean it and let me eat.

**microphone**: So you don't have to worry about it by tomorrow she could have like.

**microphone**: So.

**microphone**: Let's that.

**microphone**: And.

**microphone**: She.

**microphone**: Likes.

**microphone**: A.

**microphone**: Lot.

**microphone**: Like she's not really.

**microphone**: Yeah so I do have like.

**microphone**: A line now she's having.

**microphone**: Right so what what can that word is and she really like that you find it like change.

**microphone**: It so that's that.

**microphone**: We probably go to kitchen.

**microphone**: There are some dreams.

**microphone**: I have one.

**microphone**: She's not in.

**microphone**: So like.

**microphone**: If.

**microphone**: You want to like just try maybe like tomorrow and I'll need some food for her.

**microphone**: Then maybe.

**microphone**: So there's.

**microphone**: That and Shah wants to play with the favorite.

**microphone**: She likes to.

**microphone**: Is this opportunity.

**microphone**: Yeah no I just need to reach out.

**microphone**: Shilah loves even.

**microphone**: On like you can see.

**microphone**: The whole.

**microphone**: Time.

**microphone**: So so yeah like you're into a 30 minute series like chicken.

**microphone**: So it comes out when you first.

**microphone**: Fostered her when she's pretty until with you so yeah she is hiding but like I just came down like okay I'll do that so you'll be like this.

**microphone**: So you should be like yeah so yeah you can just land on like.

**microphone**: Yeah.

**microphone**: So I think.

**microphone**: Something.

**microphone**: Much better yeah.

**microphone**: Oftentimes you like to stay on.

**microphone**: The top of these.

**microphone**: All so yeah.

**microphone**: The names that.

**microphone**: Yeah he's pretty unique.

**microphone**: There is nobody had anything different about like two weeks you'll find something to like yeah.

**microphone**: I think maybe think of like what is she likes to answer that what are not included.

**microphone**: That.

**microphone**: So dry and then take it up for two months of prior.

**microphone**: Ity something.

**microphone**: Like.

**microphone**: Okay then okay you can leave the whole like.

**microphone**: Food I call it.

**microphone**: This is there.

**microphone**: Yeah you can see this is the thing partnership.

**microphone**: Let's figure out what I think on this bring it all with me so I'm not leaving your garbage can.

**microphone**: Worry about.

**microphone**: Yeah.

**microphone**: I have got this loop in there.

**microphone**: In this slide.

**microphone**: You don't have.

**microphone**: To generally that size which level.

**microphone**: And thank you for having this.

**microphone**: Okay?

**microphone**: So what happens basically like I check in and you get the words that I checked it you know what I mean your home.

**microphone**: Basically.

**microphone**: I.

**microphone**: Will give you like.

**microphone**: A little.

**microphone**: Write up.

**microphone**: What we did.

**microphone**: Private though I've been seeing that like university.

**microphone**: And then on the act it feels like a tiny scale of mine and then all send so you'll have like pictures to write up within the whole day.

**microphone**: Whatever.

**microphone**: We do together in a pretty straightforward and then as far as the key how am I going to get it?

**microphone**: Okay.

**microphone**: Yes so I use I've got like for you that's actually.

**microphone**: Oh okay.

**microphone**: Okay.

**microphone**: So.

**microphone**: Yeah.

**microphone**: On Saturday when you leave them on the floor I can get in with my phone okay so.

**microphone**: You can do.

**microphone**: That.

**microphone**: So you can get.

**microphone**: And like.

**microphone**: Thank you.

**microphone**: So much if you have any questions then.

**microphone**: Something that you can tell me.

**microphone**: Yeah okay no idea the feature you're like super wrong so I want to thank that again.

**microphone**: Oh okay okay.

**microphone**: Yes I yeah I have looked at that.

**microphone**: Long story of how we.

**microphone**: Decided.

**microphone**: Okay.

**microphone**: I love these buildings.

**microphone**: That have a few times.

**microphone**: Okay the same like any of like the game building has like in line over on the same yeah.

**microphone**: Perfect.

**microphone**: Perfect yeah like knowing that you.

**microphone**: Oh like I just wanted you to know I have not figured out like camera so if you have cameras in here if you want the camera like feel free I don't feel weird about that like he or her whatever like I know it's very hot and.

**microphone**: Yeah.

**microphone**: Perfect I appreciate that if we do like maker self talk you bring up for Christina yeah.

**microphone**: So you have a gap.

**microphone**: For.

**microphone**: A good five in the morning and five at night oh I think so.

**microphone**: Yeah and yeah she may like me a little bit so.

**microphone**: I feel like yeah she might feel like we both.

**microphone**: Know how to watch.

**microphone**: Yeah.

**microphone**: So that might also display into like.

**microphone**: An hour.

**microphone**: Yeah but again.

**microphone**: Thank you so much.

**system**: Ation.