The same prompt at 4.3 MP
From Qwen-Image 2.1, run locally
- Output
- 2528×1696
- Generated in
- 413.3 s
- Steps
- 40, euler / simple, cfg 1.0
- Seed
- 21
- Ran on
- RTX 3060, 12GB (local)
A deliberately texture-dense close-up, for the resolution ladder. The point of this prompt is to put as much fine structure in front of the model as one frame can hold — skin pores and freckles, fine flyaway hair in two dye colours, tattoo line work, engraved metal, knitted wool — so that a resolution that cannot resolve detail has somewhere obvious to fail. The existing R1 portrait is a good photograph but a soft one, and §36 showed the difference between 1 MP and 4.3 MP on it is real but needs looking for. Written to the PE-T2I spec: one observational paragraph, present tense, describing what is in the frame rather than instructing a camera. No text anywhere in it, on purpose. §35: a prompt that asks for text it cannot spell out turns the output into an unattributable result, and a tattoo is exactly where lettering would creep in.
