logtd

Uni-1 is here! A new kind of model that thinks and generates pixels simultaneously. Less artificial. More intelligent.

Introducing Uni-1, Luma’s first unified understanding and generation model, our next step on the path towards unified general intelligence. lumalabs.ai/uni-1

Stop guessing. Start directing. Ray3 Modify is now in Dream Machine. Edit and reimagine videos with all-new precise keyframe and character reference controls. Your vision, reimagined. Supercharge your production with rapid retouching, precise element swapping, and scene redesign.

Introducing Modify Video. Reimagine any video. Shoot it in post with director-grade control over style, character, and setting. Restyle expressive performances, swap entire worlds, or redesign the frame to your vision. Shoot once. Shape infinitely.

@roman_szczesny planning on adding it to my ComfyUI nodes for Hunyuan here github.com/logtd/ComfyUI-…

@logtdx Nice! How did you achieve it? Is there an implementation for ComfyUI?

experimenting with regional prompting on the Hunyuan video model, giving some inception vibes left side prompt: cyberpunk & pan left right side prompt: steampunk & pan right

@estebs most of them support it, but i'm not sure if anyone's implemented it before

@logtdx is this the first video model that supports regional prompting?

@asciidiego @david_kup @ErwannMillon of course diego

FlowEdit is an inversion-free method to edit images and videos. Releasing implementations of it for ComfyUI on a few models: * Flux -> github.com/logtd/ComfyUI-… * LTXV -> github.com/logtd/ComfyUI-… * Hunyuan -> github.com/logtd/ComfyUI-… (WIP) Some examples below 🧵

And last but never least, Flux. FlowEdit really shines in that it can make precise edits while keeping the majority of the image intact (a bit more difficult to pull off in video though). github.com/logtd/ComfyUI-…

LTX with STG can make some amazing changes to videos quickly github.com/logtd/ComfyUI-…

@ostrisai not yet, but hopefully soon

Has anyone had any luck converting FLUX LoRAs to SVDquant format? I have been trying to reverse engineer the process but keep hitting roadblocks.

@aindmix If you're looking for an open source Viggle this is much closer github.com/Yukun-Huang/Dr…

@logtdx So basically open source Viggle???

Just published a set of ComfyUI nodes to use Genmo's Mochi to edit videos. github.com/logtd/ComfyUI-… It uses rf-inversion, the gift that keeps on giving.

@wildfireworlds On a single 4090 it takes about 2 minutes to "warm up" a video, then 3 minutes to generate. So if you have the same video clip about 3 minutes each after warming up.

@StraughterG It might, I haven't been keeping up with Comfy's compatibility with apple silicon and the newer video models

@logtdx Yea but does it work with apple silicon

@nahbee80 if you're looking to reproduce content from one image into another, I don't think there is a good way right now if you're just looking for something similar or to remix an image, RF-Inversion is really good

@logtdx whats the best way for content transfer for flux in your opinion?

Been revisiting Reference-Only Control for Flux. It uses the diffusion model as a pseudo image encoder on a reference image to influence the generation. Results are somewhere between style and content transfer.

@natanielruizg @litu_rout_ ahaha yeah, didn't know what I was missing out on

@logtdx @litu_rout_ you’re on twitter now, awesome!

RAVE and FLATTEN were two of the papers that originally got me into diffusion models. They take inverse noise and apply consistency to image models. Now with RF-Inversion (thanks @litu_rout_ and @natanielruizg) I can try these on Flux. Not production quality, but still fun.