Image & video models: the quick picks.
TL;DR
Images: choose FLUX.3 Q4 for everyday posts: attractive, consistent and fast. Choose Qwen Turbo Q4 for portraits and photo-style posts. Q8 wasn’t visibly better.
Video: LTX-Video 2B Q8, text to video is the practical winner. It is much faster than Wan or Cog, but still needs checking for motion and anatomy flaws.
| Model | Time shown | Quality /100 |
|---|---|---|
| Images | ||
| FLUX.3 Klein · Q4_K_M | 45s | 87 out of 100 |
| Qwen Image 2.1 Turbo · Q4_K_M | 3m 22s | 88 out of 100 |
| Qwen Image 2.1 Turbo · Q8_0 | 7m 33s | 88 out of 100 |
| Qwen Image 2.1 · Q4_K_M | 6m 33s | 84 out of 100 |
| FLUX.1 dev · Q4_0 | 9m 20s | 80 out of 100 |
| Qwen Image 2.1 Turbo · Q2_K | 3m 13s | 75 out of 100 |
| Videos | ||
| LTX-Video 2B 0.9.8 · Q8 · text to video | 1m 46s | 65 out of 100 |
| Wan2.1 1.3B · Q8_0 | 34m 16s | 50 out of 100 |
| LTX-Video 2B 0.9.8 · Q8 · image to video | 1m 45s | 45 out of 100 |
| Wan2.1 1.3B · BF16 baseline | 1h 04m 06s | 35 out of 100 |
| CogVideoX 2B · Q8 · corrected integration | 40m 40s | 25 out of 100 |
| MiniMax H3 · W6A8 | No output | — not scored; no output |
Times: measured worker duration for the shown example, including loading and overhead. Scores: editorial visual quality across reviewed outputs, not a matched benchmark. “—” means no output to judge.
Image examples
FLUX.3 Klein · Q4_K_M
45s
Qwen Image 2.1 Turbo · Q4_K_M
3m 22s
Qwen Image 2.1 Turbo · Q8_0
7m 33s
Qwen Image 2.1 · Q4_K_M
6m 33s
FLUX.1 dev · Q4_0
9m 20s
Qwen Image 2.1 Turbo · Q2_K
3m 13s
Video examples
LTX-Video 2B 0.9.8 · Q8 · text to video
1m 46sWan2.1 1.3B · Q8_0
34m 16sLTX-Video 2B 0.9.8 · Q8 · image to video
1m 45sWan2.1 1.3B · BF16 baseline
1h 04m 06sCogVideoX 2B · Q8 · corrected integration
40m 40sMiniMax H3 · W6A8No completed output to show or score.