RT by @koguGameDev: ゲーム向け生成AIのScenario @Scenario_gg にも協力頂いていて参加者は期間中APIなど使えるクレジットが無料で使えます！ その他のAIツールもサポートいただける予定で交渉中です。是非ご参加ください！

RT by @gooramaru: お手伝いしている @Witchpot_ さんが、生成AI使ったゲームジャムをオンラインで開催します。生成AIを少しでもゲーム制作に使えば参加可能で、ノウハウ提供や技術相談、期間中 @Scenario_gg という2Dアセット生成サービスを無料で使えるなど参加特典も。是非ご応募ください。

R to @araminta_k: Ok I look forward to checking it out. The moment someone comes up with the easy to use and trouble-free system that I described, with the least amount of postwork required by me to clean up the art, I'm all in. I suspect a truly seamless workflow is still a few years away tho.

R to @Tom_TwToronto: It wouldn't be possible to do this with DallE-3, simply because they don't allow finetuning in a way that could learn a character to that degree - you will still struggle with consistency issues with most zero shot methods. Our next release is much closer to this workflow.

R to @Tom_TwToronto: It works on SDXL. Dalle3 is great, but can’t be finetuned so far

R to @Tom_TwToronto: We have the same type of “Controlnet mixes” in the web app (4 different modes). And more combinations possibles in the API.

R to @Tom_TwToronto: I haven't used Scenario's ControlNet in a few months but I know PromeAI's ControlNet let's users use multiple Controls all at once (Pose, Canny, Depth) and set the weights of each Control. They are lacking a control though for face and for preserving colors however.

R to @Tom_TwToronto: but from what I know, ControlNet right now works with SD1.5. Ideally it should be implemented for a really well-trained model like DALLE-3 which so far I think is the best at doing generative AI art. A ControlNet for faces specifically would then preserve facial features best.

R to @Tom_TwToronto: AI would then pose the character according to the ref photo, then render it out in whatever style I choose (be that a Scenario pre-fab style or my own trained style). I think that's what everyone is waiting for. ControlNet has been a huge step towards that (pose, canny) 2/?

R to @araminta_k: The idea situation for me as a creator, is I give the AI a character sheet (either generated by AI or drawn or as a 3d character) so that it learns what it looks like from all angles. After that I would simply give it poses via reference photos (or 3d skeletal poses) and the 1/?

R to @araminta_k: So if you compare the image you posted with the original render, it would be things like eyebrow weight, nose shape, lip thickness, eye shape. Notice your version captures the long face but lost her facial expression (happy, soft smile). The hair is close, but still not the same.

R to @Tom_TwToronto: I'll add North American comics to the request list :) I do think you'll find that with a small dataset of your character, you'll be able to get what you are aiming for.

R to @araminta_k: Yes, that's obviously much closer to the original render. For a more north American comic book style, the inklines would ideally be more black and the inkline's width would vary in size depending on which part of the body is being inked (any superhero comic book will demo this)

R to @emmanuel_2m:

R to @emmanuel_2m: Indeed, there should be ways to reasonably teach the AI both "who" the character is and transpose that to another style. Because we don't have a dataset for the character in this case, it will be naturally a bit more general :)

RT by @the_scholar1: Explore enhanced control over your creative workflows with our newest Canvas on Scenario! Hone your artistic vision and produce premium assets for your games 👾 Access now via our web app and API 👀