![Cutscene Artist](/img/default-banner.jpg)
- 73
- 134 077
Cutscene Artist
United States
Приєднався 15 кві 2017
real-time animation, 3D figure creation, and generative storytelling.
CutsceneArtist.com
CutsceneArtist.com
Offline AI Voicebot in Unity
I made an animated chatbot with Unity and Ollama. It's completely offline and hosted locally with no subscriptions.... So why is Llama 3.1 trying to convince me the AI is a spy?
(Because Meta is trash, that's my guess.)
(Because Meta is trash, that's my guess.)
Переглядів: 554
Відео
Msty: Local AI Simplified
Переглядів 1,2 тис.Місяць тому
Msty.app makes it easy to host language models on your computer or local network. 0:00 Msty.app 1:20 Live Web 2:12 Parallel Chats 2:45 Prompt Library 3:20 Knowledge Stack 4:00 Attachements 4:43 Ollama Service 5:40 Apple Shortcut
DRAW THINGS Detailer for Faces
Переглядів 1 тис.Місяць тому
Detailer scripts are now in the community cloud scripts in DRAW THINGS, the Stable Diffusion client for Mac and iOS. 0:00 Detailer scripts 1:44 Single Detailer 3:32 configure Detailer 5:44 Mask Blur 7:50 free Detailing
DRAW THINGS Model Mixing in Stable Diffusion
Переглядів 481Місяць тому
Model Mixing options in DRAW THINGS - the Stable Diffusion client for Mac and iOS. I'm covering how to bake a model with a stack of LoRAs, including converting a regular model into an accelerated model. Merge two Stable Diffusion models safely using Weighted Sum, and transfer Dreambooth training from one model to another using Add Difference. Before we're done, I'll show how to Quantize the mod...
DRAW THINGS Photo to Art - Stable Diffusion on iPhone
Переглядів 1,5 тис.Місяць тому
Generate AI art from a photo with Stable Diffusion running locally on iPhone. Tutorial for IP Adapter Plus Face, an add-on for Stable Diffusion that creates AI art from selfies. I use iPhone and the app DRAW THINGS, a free client for Stable Diffusion on Mac and iOS. Basic setup and parameters are included. Suitable for beginners new to Draw Things (but familiar with image generation). CHAPTERS ...
DRAW THINGS Accelerate Stable Video Diffusion
Переглядів 5 тис.2 місяці тому
Accelerate Stable Video Diffusion in DRAW THINGS for Mac and iOS. 00:00 AnimateLCM-SVD 01:28 Draw Things 04:08 Start/End Frame Guidance 05:16 Frames & FPS 06:17 Motion Scale 08:05 Export
DRAW THINGS Stable Diffusion 3 on iPhone
Переглядів 6232 місяці тому
SD3 has been quantized in DRAW THINGS to run on my iPhone!
DRAW THINGS Accelerate with 8-Step Renders
Переглядів 2,1 тис.2 місяці тому
Accelerated Stable Diffusion models that render in less than 10 steps, in DRAW THINGS for Mac and IOS. 00:00 Accelerated SD 01:54 fast XL LoRAs 02:58 settings and samplers 04:47 DT under 10 steps 05:45 image adjustments 07:00 workflow
DRAW THINGS Inpainting and Outpainting
Переглядів 2,1 тис.2 місяці тому
Learn how to Inpaint and Outpaint in DRAW THINGS for Mac and IOS 00:00 Fooocus Inpaint LoRA 04:10 Outpainting
Draw Things: OpenPose Control for XL
Переглядів 1,2 тис.2 місяці тому
Add OpenPose XL to DRAW THINGS... Thibaud's OpenPose SDXL: huggingface.co/thibaud/controlnet-openpose-sdxl-1.0/tree/main 00:00 Download and Import 02:49 Extract Pose / Paste to Pose 05:46 edit pose 07:11 problems 08:47 Save 09:36 Scripting
DRAW THINGS: Transparent Backgrounds
Переглядів 1,4 тис.4 місяці тому
DRAW THINGS, Stable Diffusion for IOS and Mac, new tools for transparency and background removal 00:00 Transparent LoRA 02:45 Erase Background 03:52 Mask Blur Outset 05:00 Depthmap Issues 06:38 Paint Background
DRAW THINGS: Import ControlNets for XL
Переглядів 3 тис.6 місяців тому
Import SDXL ControlNets from Diffusers and Stability AI Draw Things for Stable Diffusion is free for Mac and IOS. 00:00 Import ControlNet for XL 03:37 extract depth map 04:18 Depth comparison 07:35 Canny comparison 10:06 Sketch 10:42 Recolor
DRAW THINGS: New in 2024
Переглядів 11 тис.7 місяців тому
DRAW THINGS for Mac and IOS has a new look and many new features! It's FREE on the App Stores! 0:00 Machine Settings 1:00 Font Size 1:30 Project Manager 2:55 Thumbnail History 4:20 Menu Rail 5:53 Prompts and Tokens 7:55 Style Templates 9:25 Import/Export 11:10 Canvas Layers 11:58 Moodboard 13:00 Canvas Tools 14:45 Parameter Settings 15:45 Model Directory 16:30 Model Manager 17:25 LoRAs 18:00 Co...
Train XL LoRA on Mac & IOS with DRAW THINGS
Переглядів 9 тис.10 місяців тому
Create/export STABLE DIFFUSION LoRA with DRAW THINGS.
Draw Things - Stable Diffusion Scripting API
Переглядів 3,1 тис.Рік тому
HTTP API and Javascript Presets! Stable Diffusion for Mac and ios gets a bit smarter! DRAW THINGS discord discord.gg/Q3h4kvuqv8 Pulp Novel Art Generator discord.com/channels/1038516303666876436/1148028889461968937 Wildcards Script discord.com/channels/1038516303666876436/1147200887782133832 0:00 API & javascripting 2:00 Pulp Novel Generator 4:15 install scripts 5:56 Wildcards script 8:00 save y...
Unlock Maximum Detail - SDXL in DRAW THINGS
Переглядів 2,3 тис.Рік тому
Unlock Maximum Detail - SDXL in DRAW THINGS
DreamLook AI - SD XL Dreambooths in about 10min
Переглядів 1,2 тис.Рік тому
DreamLook AI - SD XL Dreambooths in about 10min
SDXL Deep Dive: high-res images for DRAW THINGS
Переглядів 4 тис.Рік тому
SDXL Deep Dive: high-res images for DRAW THINGS
Ai writes 'Erotic' Roleplay Prompts...?
Переглядів 663Рік тому
Ai writes 'Erotic' Roleplay Prompts...?
DreamLook AI - train dreambooth in minutes
Переглядів 1,2 тис.Рік тому
DreamLook AI - train dreambooth in minutes
Live AI CO-HOST in Unity with ChatGPT and ElevenLabs
Переглядів 455Рік тому
Live AI CO-HOST in Unity with ChatGPT and ElevenLabs
INKY: create WEB COMICS and INTERACTIVE STORIES
Переглядів 522Рік тому
INKY: create WEB COMICS and INTERACTIVE STORIES
Wrinkle Maps in iClone 8.2 - Character Creator 4.2
Переглядів 541Рік тому
Wrinkle Maps in iClone 8.2 - Character Creator 4.2
DreamBooth Virtual Actors: train AI Models from 3D figures
Переглядів 682Рік тому
DreamBooth Virtual Actors: train AI Models from 3D figures
you can set the system prompt to make the bot act however you want. if you want him to be positive but it in the system prompt
Also, any word on a controlnet for SD3 to do IP Adapter Plus Face?
IP Adapter FaceID for Kolors should be released in a few days (currently beta testing)...
what is the "zavy-ctflt" part of the text prompt?
Zavy is on Civit.ai, makes great art LoRAs. the last bit is the trigger for this particular LoRA called 'cute flat' ... a search should bring it up on Civit.
Is this "CosmicDynavision" model one you installed or mixed yourself? I can't find it but it looks great. Thanks for the help. I'm messing with SD3 and the initial output was... blobs... but after watching this and another tutorial the output is MUCH better (albeit not as good as yours yet)
It is a mix of Cosmicman and DynaVision... I have a recent video about model mixing in DrawThings.
03:17 what is that style of photos called? which prompt did you use for it? i’m searching for it a long time. would really appreciate an answer 😁
you are my Savior Angel... God Bless You 🥺🙇♂
this was amazing! can you also make one about uxing fux dev with realism_lora? would be so good
привет! классная аватарка 🤝
I use macbook air M2 why i can’t run it ?
Bravo!!! <3 fell immediately for it! I am trying to do a speech-2-speech pipeline with a Vanilla frontend to a mlx-python backend that uses coquiTTS, mlx whisper and mlx llama3.1 models, as well as VAD for word recognition, but my struggle comes with lip sync, although now the idea of a robot it's delightful to avoid the hassle. I do have a couple of options on the test table if you'd like to exchange ideas! Love your channel
Yeah! I'd love to hear what you are trying!
You mentioned watching the previews to see if an image is being overcooked and maybe you don't need as many steps to create an image? Does that mean you need to sit there and count how many preview images appear before you get a good image or does it tell you somewhere in the interface which step you're currently on? Thank you.
You don't need to count steps. It is an objective way to know if/when the render is all the way 'cooked'. But like grilling a steak, sometimes you prefer it a rare or prefer more charred... Parameters will be different for every model, every prompt (character length), and whether you want a subtle realistic 'photo' or a bold illustration. It is better to learn when to manipulate the controls than to leave every render at same generic settings. You don't need to guess an exact parameter, it's enough to understand when a setting needs 'more' or 'less'..., that's enough to go in the right direction. It is an art, like photography. Your 'eye' will develop as you spend more time with a specific model, and learn where its sweetspots are for your ideas.
Great work... I've been using whisperv3 large and xtts2 with llama 3.1 8b, I found ollama to have inferior model versions than the original, because the 16 bit model was not the same quality as using something like vllm 16 bit, but that's because my chatbot keeps its own memory through prompting, which requires low hallucinations. I'm going to check out unity assets man, thanks! Try llama 3.1 storm 8b through vllm, make your own endpoint.
Wow, thank you for suggestions! I know I need to spend some time trying different models.Your tips are appreciated!
@@CutsceneArtist cheers!
AI is so crap 🤣🤣
🤣 FACTS!
Really nice video, well done! Mind if I ask which Mac you are using there?
I have an M1 Mac Studio, 64GB
@@CutsceneArtist Thank you. Can I ask if there generations in this video were sped up at all or are they realtime? Thank you.
I sped up the video to fit the narration.
@@CutsceneArtist Ah okay thank you. I thought it was a bit fast even for a Mac Studio ;-)
This is looking good!
Thank you!
Fun video! What a magical combination of tech. So many possibilities...
niiiiiiiiice 😍
Slime is the best and cheapest
Is the audio out of sync for anyone else about halfway through the movie? Nothing she's saying is lining up with what she's doing? Seems to be about 30 seconds out 😞
I messed up the narration, sorry.... The description and the video are both correct, but you are correct, I am not showing the steps as they are talked about.
@@CutsceneArtist Ah thank you. Thought I was just tired when I was first watching it ;-) No problem at all. A very easy thing to do. Thank you for the fantastic videos though. All the very best.
Looks amazing! What do you mean by overcooked and how are you checking this? Do you have a video explaining this?
Thank you soooo much for all your content. As Mac user I am really thankful for your great videos.
Great primer - best I've seen. Nice one!
your videos is so helpful. i wonder what type of controlnet preprocessor of sketch r256 to select when importing?
Sketch and Recolor set the preprocessor type as INPAINT. (It is not at all obvious they would be inpaint)
EDIT: UPGRADING THE MAC OS FROM VENTURA TO SONOMA SEEMS TO HAVE FIXED THE ISSUE. Drawthings gave about 10 great Flux model outputs and then the program quit unexpectedly. Restarted and now it only gives outputs of greyed out blocks. Restarted computer. Reset default settings. Still grey blocks. Is this a common issue with a simple fix? This is the first program I have ever had be permanently disabled by a crash.
I'm not the dev. Try deleting the file from your application folder and redownloading (all the models and project history is in your root drive and won't be erased) If your copy of flux was imported through a HF download, you may need to trash it, and try one of the internally-downloadable versions so it gets imported with any developer bells and whistles pre-programmed
@@CutsceneArtist Thanks for the quick response. Is this a common and well known issue with a known cause that I can avoid in the future? I only had about 28GB free space on my hard drive. Was thinking maybe that was the issue.
It's not a known issue to me... the part where it worked and the stopped..., idk that sounds like a memory leak or something? I guess standard advice: make sure you have the latest version of DT.
@@CutsceneArtist How can we get access to the discord? All the links I find on reddit are expired. Maybe they can help with my issue. I deleted the app and reinstalled but I can tell it wasn't a true delete as it immediately brought up my previous settings and the problems persist. Seems very strange that a single crash can permanently disable a program that had been working properly. Such a shame too as it is an amazing tool.
this link should never expire: discord.gg/Q3h4kvuqv8
Cheers for such a helpful tutorial, had tried to do a local install of FLUX on my Mac M1 but could not get it to work. Then I implemented the use of Hyper 8 Lora with FLUX and to my amazement it works rather well with 10 steps! Still takes around 10 minutes per render but quality is great : )
could a new model be mixed which excludes all cartoon anime results, and only retains photo quality results in the new model?
I would prompt for 'cinematography of...', and don't go too high with CFG/Text Guidance. There are a few cinematography models I like that have been trained on photoreal imagery (presumably movie stills).
V14 and Florence 2 (in comfy ui ) seem to be a go to place for image to prompt good results.
can Claude and Grok be used ?
I think so...? I haven't tried, but I assume they all use similar API...
What are you using to make a talking avatar
this was iClone
That's neat. How do you use it with local models either loaded from e.g. KoboldAI or Ollama/LlamaCPP which seem to have their own web clients?
Unfortunately these clients don't share their models very well.... I think Msty can use Ollama models, but otherwise it's another app taking up enormous amounts of hdd.... (I am trying not to download a bunch, lol)
Using apple shortcuts to connect this and SD sounds amazing. How did you achieve that?
I was struggling with API from Unity..., but Shortcuts is one of those things where fewer options makes it easier to set up (harder to do it wrong). Maybe needs a tutorial video?
@@CutsceneArtist Yes, please. I could need also a tutorial on Apple Shortcuts. Find it hard to understand, even after watching some tutorials.
You mentioned that you found an alternative to Whisper for STT. So what was this alternative, if I may ask?
Recognissimo, in Unity asset store
Oh wow,.. this is amazing! And with out of the box support for your Obsidian vaults!
Another great video! I really like the Draw Things music that you made!.Great stuff!
Oh wow this is going to make such a difference when I can finally get my hands on this stuff!
HEP!!! <3
@@CutsceneArtist it’s been a bit. Sorry for that. I still watch the videos and thumbs them up but I forgot that thumbs are basically anonymous and if I want you to know I’m watching it I need to drop you a line. 😅
very cool,...have you also made the experience that when in auto1111 certain settings create real 100% photos, and where absolute identical settings etc in Draw things create cartoon images ?
I never used A1111.., but models looking more 'cartoon' makes me think maybe the CFG (text guidance in DT) is set too high, or maybe the sampler is overcooking.
THIS is what I’m looking for! Thank you!
I tried IP adapter face plus sd 1.* with generic stable diffusion 1.5 model and it's not working at all if I use the Hyper 8-step lora. Without lora, it's working... I don't know what to do 😮💨
Ahh, it sounds like the acceleration LoRA is throwing off the schedule for the Control... It might help to lower the LoRA weight, or to boost the Control weight over 100%... or maybe remove the LoRA once you have a good idea how it should look and suffer through a long render for the 'beauty shot'...
Thanks for your continued hard work!
Good video
Another great video on Draw Things!
Liu Liu just dropped Kwai Kolors XL model on Draw Things and it is amazing! It works so well with prompts and TCD LoRa and sampler!
Amazing!! This is the tutorial I have been waiting for. Thank you ❤
Let me know how it goes!
@@CutsceneArtist i went great on my ipad pro m4 1tb model, thanks for the help
很喜欢你的讲解,希望能多一点教程给我们新入手的朋友
Yeah!!
Fiiiiinalyyyyyyy! Thanks for the tutorial!
Agreed
Wow! Thanks! I can learn so many things from you! :D
Balenciaga
THANKS !
On version v4.41 (4.41.2812.1) released on Apr. 17th, 2024 they added: "The Export FBX panel now has the option to export character expressions as pure morph data, including "Jaw Open", "Jaw Left", and "Jaw Right"." But I can't find that option in the FBX Export Panel so far. Have you seen it by any chance?
i will look
@@CutsceneArtist I figured. We still need to do the steps in this tutorial, but additionally, we need to check the box "Mouth Open as Morph" in the Export Panel under the gear icon. And we will get the same visemes/blendshapes 3 times generated, example: Ah, Ah.001 and Ah.002. The first original one is bone dependent, the second is bone free blendshape and the 3rd is completely useless. That for each visemes or expression. Basically now we gotta do extra steps and have to delete useless blendshapes. Way to go Reallusion :P
@@NexVice OOF! Thanks for following up...
awesome ❤