2025 has seen further acceleration in open source model releases and incredible breakthroughs.
Z-Image Turbo by @Ali_TongyiLab recently dropped and I am blown away by how good this model is.
@SD_Tutorial It's a banger in instruction following and realism and has far better control of composition.
High IP knowledge and the dreaded baked in safety filter became a non-issue with 0.1-0.4 nudge from certain edgy LoRAs (no text enc, layer multi, etc. hacks)
https://t.co/ZEfAopmijt
@berkaykilic_2@SD_Tutorial I thought the same but work back from a crazy good image that contains a (clearly working) good workflow.
This model trades blows with closed source massive models and the safety false positives go away entirely.
And it seems to know all IP
@berkaykilic_2@SD_Tutorial I thought the same but work back from a crazy good image that contains a (clearly working) good workflow.
This model trades blows with closed source massive models and the safety false positives go away entirely.
And it seems to know all IP
@andreintg Killer feature would be ability to pull the generated shape from the sync camera with something like tripo (or another image-to-3D backend) with some (admittedly painful) magic, this would be insane.
@marshall_a60950@brave Omg you are right it's trivial to get around.
Get a handle to the player and just:
vid.playbackRate = 3;
Hahaha awesome..
Personally I'm just going to write a userscript to add the buttons I want (3x and 4x) to the native UI so its clean.
@ti0719 This would serve as an excellent benchmark for humanoid robots. Fine hand and finger control working with malliable and delicate materials, fine detail painting, bending, stretching, intricate tool use, etc. With an easy to score result (likeness to pro-made from many angles)
PSA: Do not open your windows in Waymos.
I was assaulted and robbed in a Waymo in the Mission District.
I was punched three times in the face and head. Waymo treats criminals as pedestrians and stops moving, leaving you vulnerable.
They gave me 5 free rides tho
@retrokafalar@BrentLynch I'm getting hard grok music vibes, but 1.5 just came out... Couldn't possibly be 2.0 already? (I mean it was teased as Imagine 2.0 coming soon, but surely not)
If so, that is a considerable leap in video prompt adherence and verisimilitude (just not audio).
@retrokafalar@BrentLynch I'm getting hard grok music vibes, but 1.5 just came out... Couldn't possibly be 2.0 already? (I mean it was teased as Imagine 2.0 coming soon, but surely not)
If so, that is a considerable leap in video prompt adherence and verisimilitude (just not audio).
@ComfyUIWiki Hey, have you considered a (albeit slightly modified) version of the official system prompt in that workflow? I swapped it out and seem to get far better results YMMV.
I did have to pass in the aspect ratio, but every thing else is much the same.
https://t.co/g3GiiSs5vA
@ComfyUI Happy I was wrong on this one.
Nice @ComfyUI!
More open weights models support with serious horsepower. Will be interesting to see how Cosmos 3 quants/nano stack up to this and the current best open models.