H3-Image at 16MP: I reran the same 15 prompts, and the gap narrows
retained text
A few people in the last thread pointed out I ran H3 at \~4MP while MOSS got its full 4MP, and that H3 apparently wants 16MP before downscaling. Fair call, so I reran it. Same 15 prompts and aspect ratios. H3 generated at 16MP (4096x4096 up to 6272x2688), then downscaled to the same display size as MOSS, so nothing here is being compared on pixel count. MOSS is the quality tier at its 4MP ceiling. What changed: Faces and skin: the 16MP renders keep pores, freckles and stray hair. At 4MP those were smoothed away. Biggest difference. Fine texture (foam, droplets, fabric): same direction, more micro-detail survives. Composition, prompt adherence, text rendering: roughly unchanged. UI mockups and charts: still fail. Resolution doesn't fix semantics. So: at 4MP I underrated H3. At 16MP it's much closer, and on skin and material detail it's ahead of what I posted before. MOSS still reads better to a lot of people in the comments, and I get why. Prompts are the same 15 as the first post, same order. First set, H3 at its default \~4.18MP: [https://www.reddit.com/r/StableDiffusion/comments/1xuft8g/](https://www.reddit.com/r/StableDiffusion/comments/1xuft8g/)
document fetched 2026-11-14 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xutwi3/h3image_at_16mp_i_reran_the_same_15_prompts_and