Qwen3-TTS, a series of powerful speech generation capabilities
https://redd.it/1qjuebr
@rStableDiffusion

PersonaPlex: Voice and role control for full duplex conversational speech models by Nvidia

https://redd.it/1qjtpf1
@rStableDiffusion

LTX2 issues probably won't be fixed by loras/workflows

When Wan2.2 released the speedup loras were a mess, there was mass confusion on getting enough motion out of characters, and the video length issues resulted in a flood of hacky continuation workflows

But the core model always worked well: it had excellent prompt adherence, and it understood the movement and structure of humans well

LTX2 at its peak exceeds Wan, and some of the outputs are brilliant in terms of fluid movement and quality

But the model is unstable, which results in a high fail rate. It is an absolute shot in the dark as to whether the prompts will land as expected, and the structure of humans is fragile and often nonsensical

I'll admit LTX2 has made it difficult to go back to Wan because when it's better, it's much better. But it's core base simply needs more work, so I'm mostly holding out for LTX3

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

Quick 1080p test on new laptpop !

https://redd.it/1qk0qyc
@rStableDiffusion

Ltx 2 music video

https://redd.it/1qjxu2a
@rStableDiffusion

AI girls flodding social media, including Reddit

Hi everyone,

I guess anyone who has worked with diffusion models for a while can spot that average 1girl AI look from a mile away.

I'm just curious by now how do you guys deal with it? Do you report it or just ignore it?

Personally, I report it if the subreddit explicitly bans AI. But Instagram is so flooded with bots and accounts fishing for engagement that I feel like it's pointless to try and report every single one.

https://redd.it/1qk0vac
@rStableDiffusion

According the Ace Step team, ACE STEP 1.5 music model releases soon! This is what I got when I asked it for a mix of dubstep, arpeggios, gritty bassline, female singer, melodic. Nice mashup IMO.

https://redd.it/1qk2odv
@rStableDiffusion

PromptBridge-0.6b-Alpha: Tiny keywords to full prompt model
https://redd.it/1qk88om
@rStableDiffusion

We are very very close, I think!
https://redd.it/1qkb160
@rStableDiffusion

How to render 80+ second long videos with LTX 2 using one simple node and no extensions.

I've have amazing results with this node:

Reddit: Enabling 800-900+ frame videos (at 1920x1088) on a single 24GB GPU Text-To-Video in ComfyUI

Github: ComfyUI\_LTX-2\_VRAM\_Memory\_Management

From the github repo:

"Generate extremely long videos with LTX-2 on consumer GPUs

This custom node dramatically reduces VRAM usage for LTX-2 video generation in ComfyUI, enabling 800-900+ frames (at 1920x1088) on a single 24GB GPU. LTX-2's FeedForward layers create massive intermediate tensors that normally limit video length. This node chunks those operations to reduce peak memory by up to 8x, without any quality loss."

This really helps prevent OOMs, especially if you have less VRAM.

You can add this node to any existing LTX-2 workflow, no need to reinvent the wheel.

I just finished a 960x544 2000 frame / 80 sec. render in 17 minutes on a 4090 24bg VRAM 64 GB RAM system. In the past, there was no way I'd come close to these results. Lip-sync and image quality hold through out the video.

This project is a work in progress and the author is actively seeking feedback.

Go get chunked!

https://redd.it/1qke313
@rStableDiffusion

Flux Dev 1 Vs Z image. Unfortunately, Zimage is very poor at generating landscapes (The same applies to WAN - But it's very good with interior scenes)

https://redd.it/1qke70f
@rStableDiffusion

Any idea guys on how to generate different types of heels?
https://redd.it/1qkbwax
@rStableDiffusion

Tensorstack Diffuse v0.4.0 beta just dropped about an hour ago and I like it!
https://redd.it/1qkjxsv
@rStableDiffusion

Flux.2 / Klein Inpaint Segment Edit (edit segments one-by-one!)

https://redd.it/1qkkv6q
@rStableDiffusion

Follow up: AI 3D generation has improved a lot, better topology & texturing results

https://redd.it/1qkl12j
@rStableDiffusion