NVIDIA PiD-based img upscaler (no workflow but .py)

I've "created" a simple img2img upscaler using the FLUX2VAE-variant of NVIDIA's PiD. It's a simple python script, not a Comfy workflow.

You'll need a 24GB VRAM GPU for 1024px and 32 GB for >1024px.

https://github.com/geronimi73/3090\_shorts/tree/main/NVIDIA-PiD-FLUX2VAE-upscaler

It's stripped of all the training related stuff in the original nv-tlabs/PiD github repo. Just torch and transformers. That's how I burned my Claude Code tokens for the day.

I think the model is pretty good. Unfortunately NVIDIA once again changed their mind when it comes to license.

https://preview.redd.it/o1ko8dr7in3h1.png?width=1856&format=png&auto=webp&s=557f50b14c380ba6255acd356fdb7d26974d71ed



https://redd.it/1tp0qzx
@rStableDiffusion

Regarding Anima, can there be a site where we see the artist styles from both Danbooru and Gelbooru? I read it uses both, but I'm only seeing sites with the Danbooru artist tags, can there be one with Gelbooru too?



https://redd.it/1tp2h86
@rStableDiffusion

Anima - Goku Transformation Series: From SSJ1 to Mastered Ultra Instinct

https://redd.it/1tp671f
@rStableDiffusion

InvokeAI 6.13 just released, its largest community-driven release ever. Adds full support for Anima & Qwen Image, support for API models (like GPT Image), support for Prompt Expansion & Image To Prompt, lasso & polygon tools, overhauled docs website and more

InvokeAI no longer has a commercial entity backing its development, this release was entirely community driven by 30+ individual volunteers.

https://preview.redd.it/b1n3s1afuo3h1.png?width=2559&format=png&auto=webp&s=cd96c211b7b72f4dbba187e017a2f114512ad97f


Highlights include:

Full Support for Anima

Text to image, image to image, and LoRAs. Support was also added for the ER SDE scheduler. Improved regional guidance support and controlnet support will be added soon.


Full Support for Qwen and Qwen Image Edit

Text to image, image to image, LoRAs, reference image, regional guidance, and controlnet support.


Support for API models such as GPT Image and Nano Banana

If local models ever can't quite do what you need it to do, you can link an API key to an external API service and generate images directly in the canvas. This was originally a feature in the paid commercial version of invoke (which no longer exists) and was built from scratch for the free community edition.


Support for Prompt Expansion and Image To Prompt

Expand your prompt using an LLM such as Gemma or Qwen Instruct, or convert your image into a prompt.


New Canvas Tools (Lasso, Polygon Tool)

Last release the Text tool and Gradient tools were added. In this release, the available tools continue to expand with Lasso and Polygon tools.


Extended Multi-User Mode

Multi-user mode now supports creating private or shared boards and workflows


New Website & New Documentation Site

After the original team behind the commercial entity was hired by adobe, the website was effectively closed down. In this release, the website and documentation sites have a new coat of paint https://invoke.ai/



Full release notes: https://github.com/invoke-ai/InvokeAI/releases/tag/v6.13.0


Download: https://github.com/invoke-ai/launcher/releases/tag/v1.8.1

https://redd.it/1tp7e6w
@rStableDiffusion

DEMON: Diffusion Engine for Musical Orchestrated Noise

https://redd.it/1tpa6tj
@rStableDiffusion

Anybody knows what kind of technique is this ?

https://redd.it/1tp9cmv
@rStableDiffusion

I'm a newbie (not really). Which are your recommendations to transform sketches into images?

https://redd.it/1tpd3xb
@rStableDiffusion

Running real-time 1080p video generation and editing on your own (Dreamverse OSS release)

https://redd.it/1tpfbrl
@rStableDiffusion

Microsoft Lens - Non Turbo with 5 CFG (ComfyUI)

https://redd.it/1tpfrvt
@rStableDiffusion