Disegno_Vivo
From a drawing of an animal to a photorealistic creature that keeps the drawing's "mistakes".
The problem
Turning a drawing into a photo is easy. The hard part is not losing the mistakes: a model trained on photographs “fixes” the creature towards the shape it knows, and the result is pretty but it’s no longer the drawing. The extra legs, the oversized fin, the crooked eyes are exactly what makes a hand drawing unique.
The solution
- The drawing’s quirks are listed in words, one by one, and become part of the prompt: proportions, what’s missing, what’s extra. Never the line style — otherwise you get a drawing back.
- The drawing is generated straight into a vertical 9:16 canvas, in a single pass, ready for social media.
- From the still photo come video clips and short edits with synchronised sound.
How it’s built
- Everything runs locally on an 8 GB card, driving ComfyUI through its API with nothing but Python’s standard library.
- The method is measured by blind judges counting the preserved quirks, against a control group without the list: the written list takes the cat from 0 to 8 quirks out of 8.
- Automatic limb counting was tried with two methods and dropped because it failed calibration: counting stays a job for human eyes.
- The prompt’s opening weighs as much as any item on the list: “real animal” instead of “3D creature” is the difference between a puppet and a shark with gills.
Screenshots
