Yue GP v3: by popular demand I have added multiple generations per Genres prompt and the possiblity to generate multiple songs based on the same lyrics but with different Genre prompts. As a bonus, I have added a progression bar and an Abort button

@McQueenFu No sorry the hugging face live demo is not maintained by me.

I guess this type of mistake is the logical consequence of naming in my code my acceleration patch “Plan 9 from Deep Outer Space”

@abrakjamson well done ! very nice example

You may want to ‘git pull’ YuE GP as I have updated the repo with In Context Learning : you can now guide song generation by providing samples of an instrumental or vocal track

@abrakjamson Good timing. I have updated the app and now you can provide audio prompts if you start it with the —icl switch

@anunidentified Happy I could help

By adding just one call in the original sourcecode to the mmgp module you can run it with only 6 GB of VRAM (profile 5). Conversely, if you have 16 GB of VRAM you can turn the turbo mode (profile 1) and generate images in less than 6s !

Here is one more great application available to the GPU Poor : Hunyuan-3D-2GP ! Try this great text or image to 3D generator which can run with only 6 GB of VRAM.

@EHuanglu @_akhaliq If may consider using this low VRAM (less than 6GB VRAM) fork of the original Gradio App :

And now ladies and gentlemen please welcome Cosmos1GP one of the best image to video / video to video open source model available which can now run on consumers GPU:

So here it is: Flux Fill GP, the best iterative inpainting and outpainting tool running at last on RTX 3090 / RTX 4090. You will need at least 48GB or RAM still.

After 15 years in the black hole, at last something came out : I am proud to announce the Hunyuan Video generator for the GPU Poor (RTX 3090 / RTX 4090). Fast and light:

@cocktailpeanut I have also in stock a GPU Poor version (RTX 3090/4090) of Flux Fill that is a great tool to iteratively do inpainting / outpainting using the Flux generator

@skalskip92 @vo_d_p Are you sure you comparing it to the newly released LLaVA 1.6 ? Version 1.6 is much better than version 1.5 and certainly the best open source option today.