sarvamaigc

/

Qwen-Image-2.1-Sarvam-Tez

Like 5

Follow

Sarvam AI-GC 3

Thanks @V33rGeer , that’s a great find! I’ve added the tip about using 3 steps instead of 2 to fix text to the README, with credit to you.

Glad you’re finding the model useful. We’re continuing to train it, so keep an eye out for an improved release soon.

Well, the part that impressed me is that it could take four text boxes in sequence, all containing garbled nonsense text, and properly format them without errors with 'just' a prompt asking it to
Change the text bubbles from left to right, "ooh", "ahh", "silly text in my prompt", "what secrets might they contain"
The output had some fuzzy letters with 2 steps, but quite clear with 3 steps.

The model combines well with a different model; https://huggingface.co/congruency/Qwen-Image-2.1-turbo-mfd/discussions/1

that particular workflow is quite a mess, though it does have some interesting bits to make the setup behave itself;

though the "flow shift" value is only relevant in so far as you desire a visually noisy output (lower is noisier, higher is cleaner, there is plenty to find between 1.5 and 4.0)