niklaslenz3 image setups
| # | Setup | Seconds per image enhancer off |
Seconds per image enhancer on | My rating enhancer off | My rating enhancer on |
Overall, as of 20 Oct 2026 |
|---|---|---|---|---|---|---|
| 1 |  MOSS-Image-SFTComfyUI · INT8 ConvRot (Comfy-Org format) | 12.81 s | 12.99 s | 7.13 | 7.25 | 52.2 speed 77.5 · ratings 43.8 |
| 1 |  Qwen-Image-2.1ComfyUI · INT8 ConvRot (Comfy-Org format) | 7.64 s | 8.23 s | 6.5 | 7.13 | 52.2 speed 100 · ratings 36.3 |
| 3 |  FLUX 3 Image KleinComfyUI · INT8 ConvRot (Comfy-Org format) | 1.96 s | 2.01 s | 6.5 | 6.25 | 45.6 speed 100 · ratings 27.5 |
Overall: 0 to 100 from speed (25 %) and my ratings (75 %). Speed is the mean of the two medians: up to 10 s per image it counts 100, above that 1000 divided by the seconds. My ratings are the mean of all 8 ratings of a setup (4 motifs, enhancer off and on), minus 5, times 20 (a mean of 5 counts 0, a mean of 10 counts 100). The scale is fixed and does not depend on the other setups. The two best setups have the same overall score, so there is no pick for this class. My ratings: 0.5 to 10 in steps of 0.5, one rating per motif and enhancer mode for the 3 seeds together, by one rater (me). Click a setup for all 24 images. Seconds per image: ComfyUI execution time of one 1024×1024 image with the model already loaded (mflux generate time on the Mac); the median of 12 images.
My rating per motif
| Motif | MOSS-Image-SFT | Qwen-Image-2.1 | FLUX 3 Image Klein |
|---|---|---|---|
| Portrait | 8.25 | 8 | 7 |
| Fire breather | 8.25 | 7.5 | 7 |
| Action shot | 6.5 | 5 | 6 |
| Alien forest | 5.75 | 6.75 | 5.5 |
Mean of my ratings with the enhancer off and on, out of 10. Bold is the best rating of the row; two bold cells are a tie.
RTX 5090 32 GB · image models
MOSS-Image-SFT on one RTX 5090 32 GB
one RTX 5090 32 GB · 50 steps, CFG 3, euler / sgm_uniform, shift 5, negative prompt set, layer skip 22 on the negative branch, 1024x1024, batch 1, seeds 42 / 123 / 456
Exact command and run record · All 24 images · All raw files
Qwen-Image-2.1 on one RTX 5090 32 GB
one RTX 5090 32 GB · 40 steps, CFG 1, euler / simple, no negative prompt, 1024x1024, batch 1, seeds 42 / 123 / 456
Exact command and run record · All 24 images · All raw files
FLUX 3 Image Klein on one RTX 5090 32 GB
one RTX 5090 32 GB · 16 steps, CFG 1, euler / simple, no negative prompt, 1024x1024, batch 1, seeds 42 / 123 / 456
Exact command and run record · All 24 images · All raw files
What is like for like here
Every setup made the same 24 images: 4 house prompts (portrait, fire breather, action shot, alien forest; the full text is in prompts.json on each recipe page) × 3 seeds (42, 123, 456) × the prompt enhancer off and on, at 1024×1024, one warm-up image thrown away. The speed numbers are the medians of the 12 images of a mode. With the enhancer on, the model's own prompt rewriter wrote the text first; those texts were written once on the RTX 5090 box and reused on every box, so the enhancer-on images all start from the same text. The enhancer time is measured in its own run and is not in the image time. For FLUX 3 Image Klein the maker recommends the prompt upsampler (enhancer on); the enhancer-off images go against that advice, so both modes are shown.
The ComfyUI setups use INT8 ConvRot weights (Comfy Org format); the Mac uses mflux with its own 8-bit quantization, so the Mac recipe differs. The same seed gives other pixels on other hardware. I rate pictures I like to look at; it is one person's taste, so read the ratings as my ratings, not as a quality measurement. Image files are published with the metadata chunk removed, pixels identical; the hashes in the time files are those of the published files (png_sha256_original is the hash of the original). All images on this page are AI-generated.