Qwen Image Edit Character Consistency
While some hosted versions show very poor character consistency, the official version from Qwen Chat performs quite well.
We will need to wait for the QQUF or low-VRAM version to test its local performance.
#qwen #qwen_image #ai
Qwen Image Edit Virtual Clothing Try-On
The virtual clothing try-on feature works consistently when prompted correctly.
By using a stitched image of the person and clothing as a single input
#Qwen #AI #QwenImage
Just released a Frappe ERPNext app with some handy project management tweaks.
First feature: Task dependencies can now be set across projects : pick dependent tasks from any project, not just the current one.
#frappe #erpnext #projectmanagement
Check it out: github.com/pushakargaik...
About one word per second response speed on 8 GB VRAM.
20B model, after all. 😅
Holy Sh!t This is not a drill.
OpenAI releases open-source reasoning, agentic model.
- 120B and 20B variants
- Apache 2.0 license !!!
#openai #opensource #chatgpt
If you want an open-source alternative to Google Genie 3, these folks are building it.
I saw WAN 2.1 used somewhere, so future versions may be optimized for consumer hardware.
#genie3 #opensource
stdstu12.github.io/YUME-Project/
Prompts from the technical report qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Image/Q...
Higher steps (30) makes text better
GGUF speeds things up and is more manageable in low VRAM
Qwen Image in ComfyUI. Now re-creating some of the examples on 8GB VRAM.
Qwen Image works in ComfyUI 🥳
Apache 2.0
SOTA opensource text rendering
Initial loading and offloading time on low VRAM were too high.
#Qwen #OpenSource
When the person walks into the puddle: My mind is melting at what it takes to make this in video game & this AI model is doing it real time
Almost photoreal
Soon no amount of money you throw at building AAA games would render as good to what can be generated, as good as real life
credit: MattMcGill_
24fps generated in real time. Shadows, water reflections, & a realistic furry dog with responsive animations to player input
Its literally a"video" game
credit: jparkerholder
" #Genie3 feels like a watershed moment for world models:we can now generate multi-minute, real-time interactive simulations"
btw, had to change this line to get Chatterbox TTS running locally
Used Chatterbox TTS to clone my voice for video voice-over.
Whisper for subtitles.
Blender VSE for video editing.
End-to-end open-source pipeline in progress. 👨💻
#ai #genai #chatterbox #tts #blender
Searching for how to generate unrestricted, unlimited, local free AI video with no conditions that creates realistic, smooth, high‑quality cinematic videos?
WAN 2.2 is the true open‑source Apache licensed model that answers all of this.
#aivideo #wan
youtube.com/shorts/BwMGz...
Want to iterate faster and get better results in ComfyUI?
I just shared a beginner tip for a simple technique that makes image generation quick, consistent, and fun. Check out how using one, four, and ten steps affects the detail—great for beginners and pros alike.
Details in the article 👇🏻
Created a handy zsh function to convert any video to audio (mp3, wav, etc.) with ffmpeg in seconds! 🎬🎵 Works on Linux, Windows, and Mac.
Full guide & copy-paste code in the post 👇
Reading JLA 1997 # 1, is that Wolverine on the left and Doom on the right?
Much better IMO compared to first try. [BSKY might have compressed the video]
A touch of compositor to make the text Star Wars yellow
Realized creating an empty, parenting all logos to it, and animating the empty instead of the camera would have been simpler and provided more control for the title crawl.
