Built a local AI creative suite for Windows, thought you might find it useful
Hey all, I spent the last 6 weeks (and around 550 hours between Claude Code and various OOMs) building something that started as a portfolio piece, but then evolved into a single desktop app that covers the full creative pipeline, locally, no cloud, no subscriptions. It definitely runs with an RTX 4080 and 32GB of RAM (and luckily no OOMs in the last 7 days of continued daily usage).
https://preview.redd.it/qhvafyragdqg1.png?width=2670&format=png&auto=webp&s=a687d9c65e7ea7173bccdda426c22f590e8c2044
It runs image gen (Z-Image Turbo, Klein 9B) with 90+ style LoRAs and a CivitAI browser built in, LTX 2.3 for video across a few different workflow modes, video retexturing with LoRA presets and depth conditioning, a full image editor with AI inpainting and face swap (InsightFace + FaceFusion), background removal, SAM smart select, LUT grading, SeedVR2 and Real-ESRGAN and RIFE for enhancement and frame interpolation, ACE-Step for music, Qwen3-TTS for voiceover with 28 preset voices plus clone and design modes, HunyuanVideo-Foley for SFX, a 12-stage storyboard pipeline, and persistent character library with multi-angle reference generation. There is also a Character repository, to create and reuse them across both storyboard mode as well as for image generation.
https://preview.redd.it/ys308jnegdqg1.png?width=2669&format=png&auto=webp&s=b1b1ef23814b193ac4e95b2cac4d869d53c5bd8e
https://preview.redd.it/c4nx2gtggdqg1.png?width=2757&format=png&auto=webp&s=ea7388165fd4424acc79e5c139584e3d92a611a5
There's a chance it will OOM (I counted 78 OOMs in the last 3 weeks alone), but I tried to build as many VRAM safeguards as possible and stress-tested it to the nth degree.
Still working on it, a few things are already lined up for the next release (multilingual UI, support for Characters in Videos, Mobile companion, Session mode, and a few other things).
I figured someone might find it useful, it's completely free, I'm not monitoring any data and you'll only need an internet connection to retrieve additional styles/LoRAs.
https://preview.redd.it/4o8k2uhjgdqg1.png?width=2893&format=png&auto=webp&s=0d8957bdd382b1b942ea727884c036b8a5b004ee
https://preview.redd.it/sbxd77bqgdqg1.png?width=2760&format=png&auto=webp&s=f65a29e2d7624f3a3eb420ad64506676202ac88d
The installer is \~4MB, but total footprint will bring you close to 200GB.
You can download it from here: https://huggingface.co/atMrMattV/Visione
https://preview.redd.it/qkce1kqsgdqg1.png?width=2898&format=png&auto=webp&s=95838223b023a8eb80ad42608de7fba26da84e30
https://redd.it/1rznto9
@rStableDiffusion
Hey all, I spent the last 6 weeks (and around 550 hours between Claude Code and various OOMs) building something that started as a portfolio piece, but then evolved into a single desktop app that covers the full creative pipeline, locally, no cloud, no subscriptions. It definitely runs with an RTX 4080 and 32GB of RAM (and luckily no OOMs in the last 7 days of continued daily usage).
https://preview.redd.it/qhvafyragdqg1.png?width=2670&format=png&auto=webp&s=a687d9c65e7ea7173bccdda426c22f590e8c2044
It runs image gen (Z-Image Turbo, Klein 9B) with 90+ style LoRAs and a CivitAI browser built in, LTX 2.3 for video across a few different workflow modes, video retexturing with LoRA presets and depth conditioning, a full image editor with AI inpainting and face swap (InsightFace + FaceFusion), background removal, SAM smart select, LUT grading, SeedVR2 and Real-ESRGAN and RIFE for enhancement and frame interpolation, ACE-Step for music, Qwen3-TTS for voiceover with 28 preset voices plus clone and design modes, HunyuanVideo-Foley for SFX, a 12-stage storyboard pipeline, and persistent character library with multi-angle reference generation. There is also a Character repository, to create and reuse them across both storyboard mode as well as for image generation.
https://preview.redd.it/ys308jnegdqg1.png?width=2669&format=png&auto=webp&s=b1b1ef23814b193ac4e95b2cac4d869d53c5bd8e
https://preview.redd.it/c4nx2gtggdqg1.png?width=2757&format=png&auto=webp&s=ea7388165fd4424acc79e5c139584e3d92a611a5
There's a chance it will OOM (I counted 78 OOMs in the last 3 weeks alone), but I tried to build as many VRAM safeguards as possible and stress-tested it to the nth degree.
Still working on it, a few things are already lined up for the next release (multilingual UI, support for Characters in Videos, Mobile companion, Session mode, and a few other things).
I figured someone might find it useful, it's completely free, I'm not monitoring any data and you'll only need an internet connection to retrieve additional styles/LoRAs.
https://preview.redd.it/4o8k2uhjgdqg1.png?width=2893&format=png&auto=webp&s=0d8957bdd382b1b942ea727884c036b8a5b004ee
https://preview.redd.it/sbxd77bqgdqg1.png?width=2760&format=png&auto=webp&s=f65a29e2d7624f3a3eb420ad64506676202ac88d
The installer is \~4MB, but total footprint will bring you close to 200GB.
You can download it from here: https://huggingface.co/atMrMattV/Visione
https://preview.redd.it/qkce1kqsgdqg1.png?width=2898&format=png&auto=webp&s=95838223b023a8eb80ad42608de7fba26da84e30
https://redd.it/1rznto9
@rStableDiffusion
Flux2klein 9B Lora loader and updated Z-image turbo Lora loader with Auto Strength node!!
https://redd.it/1rztbjm
@rStableDiffusion
https://redd.it/1rztbjm
@rStableDiffusion
Reddit
From the StableDiffusion community on Reddit: Flux2klein 9B Lora loader and updated Z-image turbo Lora loader with Auto Strength…
Explore this post and more from the StableDiffusion community
Qwen 2512 is very powerful. And with the nunchaku version, it's possible to generate an image in 20 to 50 seconds (5070 ti)
https://redd.it/1s00mbg
@rStableDiffusion
https://redd.it/1s00mbg
@rStableDiffusion
Reddit
From the StableDiffusion community on Reddit: Qwen 2512 is very powerful. And with the nunchaku version, it's possible to generate…
Explore this post and more from the StableDiffusion community