Stable Diffusion 3.5 Large vs FLUX.1 dev for images
Two popular text-to-image models for a 12 to 24 GB graphics card
Facts from the reviewed catalog
| Specification | Stable Diffusion 3.5 Large | FLUX.1 [dev] |
|---|---|---|
| Task | Image | Image |
| Parameters | 8B | 12B |
| License | Publisher terms: Stability AI Community License | Publisher terms: FLUX.1 [dev] Non-Commercial License |
| Commercial use | Allowed with conditionsRegister with Stability AI; organisations with over USD 1M annual revenue need an enterprise license · license text | Not allowedlicense text |
| Quality · LMArena 2026-10-07 · text-to-image | 938 ± 5934–943 · 23,396 votes · rank 81 · source | Arena rated the FP8 build (flux-1-dev-fp8); this entry is the original FLUX.1 [dev] weights · source |
| Memory | publisher's documented path from 11 GB (CUDA, conditional) | labelled 16-bit estimate (~31 GB), no documented minimum |
| Checkpoint | official card | official card |
Only Stable Diffusion 3.5 Large has a result in LMArena 2026-10-07 · text-to-image (938 ± 5); FLUX.1 [dev] has none, so the two are not ranked from this source and no score is inferred for it. Blind pairwise votes by people on hosted outputs, CC BY 4.0. A rating is relative; overlapping intervals mean the order is not settled. Scale: Arena rating ± half the confidence interval.
Memory on three machines
Each machine's own reserve is added (graphics cards 2 GB; unified and system memory 25%, 2 to 8 GB). Planning estimates, not measured runs.
| Machine | Stable Diffusion 3.5 Large | FLUX.1 [dev] |
|---|---|---|
| RTX 3060 · 12 GB VRAM | 11 GBWithin a conditional path (offload or quantization) | ~31 GBLess memory than the 16-bit estimate |
| RTX 4060 Ti · 16 GB VRAM | 11 GBWithin a conditional path (offload or quantization) | ~31 GBLess memory than the 16-bit estimate |
| RTX 4090 · 24 GB VRAM | 11 GBWithin a conditional path (offload or quantization) | ~31 GBLess memory than the 16-bit estimate |
Stable Diffusion 3.5 Large: publisher's documented path from 11 GB (CUDA, conditional). FLUX.1 [dev]: labelled 16-bit estimate (~31 GB), no documented minimum. Formulas
Everything that runs on each machine: RTX 3060 · 12 GB VRAM · RTX 4060 Ti · 16 GB VRAM · RTX 4090 · 24 GB VRAM
Which to choose
The licence is the first difference. FLUX.1 [dev] is non-commercial; Stable Diffusion 3.5 Large allows commercial use if you register with Stability AI and stay under USD 1 million in annual revenue. FLUX.1 [dev] is larger (12 billion parameters against 8) and the catalog has no documented memory path for it, so its rows use a labelled 16-bit estimate; Stability documents an 11 GB conditional path for SD 3.5 Large on CUDA. Only SD 3.5 Large has a rating in the LMArena text-to-image release.
Files and run commands
The default file of the most downloaded collected source, or the commands each card documents. Not run by this site.
Stable Diffusion 3.5 Large
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
ComfyUI · ComfyUI official examples · SD3.5: ComfyUI workflow
FLUX.1 [dev]
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
ComfyUI · ComfyUI official examples · Flux Dev: ComfyUI workflow
Numbers each publisher reports
Each card's own results with its own settings. Publishers choose different tests and settings, so these values are never compared across publishers here.
Stable Diffusion 3.5 Large · reported by Stability AI
FLUX.1 [dev] · reported by Black Forest Labs
Limits and dates
The published quality benchmark does not test a local quantized build. Memory estimates are not speed measurements or guaranteed hardware fits. Repository size, when available, is not VRAM. Formulas
Page updated · Catalog reviewed · LMArena 2026-10-07 · text-to-image