Contains affiliate links. We may earn a commission — it never changes the verdict.Details
The Stack / Image
Stable Diffusion
03 CreationYes for builders who want image generation they own — Stable Diffusion's open weights run locally, fine-tune freely, and power a massive LoRA/ComfyUI ecosystem. The current open flagship is still the SD 3.5 family; Stability's 2026 releases have gone to audio (Stable Audio 3.0) and its Brand Studio platform, not a new image generation. Correction: this page briefly claimed an April 2026 "SD4" release — that claim came from fabricated third-party articles and Stability has announced no such model.
The job in the stack
The self-hosted image seat: open-weight generation with total control, zero marginal cost, and the largest tuning ecosystem.
Pricing
| Plan | Price | Notes |
|---|---|---|
| Open weights (local) | $0 | SD3.5, SDXL, SD1.5 on your GPU |
| Stability API | $0.03–$0.08/image | Credits at $0.01 each; Stable Image Core $0.03, Ultra $0.08, SD3.5 Medium $0.035 |
Pricing verified 2026-08-26. Plans change — the source of truth is Stable Diffusion's own pricing page.
Models & versions
- SD3.5 (Large / Large Turbo / Medium)October 2024 — the current open flagship family; TensorRT-optimized builds (June 2025) run 2x faster on RTX GPUs
- SDXL / SD1.5Older open releases still everywhere in practice — the ComfyUI/A1111 ecosystems and the bulk of the LoRA library are built on them
The honest read
What works
- +Fully open: local, private, unlimited generation
- +Unmatched community ecosystem of fine-tunes and workflows
- +The customization ceiling — train exactly your style
What breaks
- −No new image generation since SD3.5 (Oct 2024) — Stability's 2026 energy has gone to audio and enterprise, so FLUX leads on raw model progress
- −Local setup demands GPU + tooling comfort
- −Stability AI's commercial licensing tiers need reading
Use it for
- →Privacy-bounded generation that can't leave your machine
- →Custom-trained styles and characters at zero marginal cost
- →Builders embedding image gen without per-image fees
Skip it if
- ×You want an app, not a model — Leonardo/OpenArt serve SD-class models hosted
- ×Peak quality per prompt — that's Midjourney or FLUX
Questions operators ask
?Stable Diffusion or FLUX?
FLUX leads on out-of-box photorealism, prompt adherence, and in-image text — and with no new SD image model since SD3.5, that lead is growing. Stable Diffusion wins on ecosystem depth: the tutorials, LoRAs, and workflows accumulated since 2022 mostly target SDXL and SD1.5. Pick SD for a tuned, owned pipeline on proven tooling; pick FLUX for the best open-weight output today.
What changed
- Freshness sweep: no changes. Re-checked Stability AI's news/updates page directly — still no new image-generation model since SD 3.5 (Oct 2024); the correction below holds.
- CORRECTION: the July 9 entry below reported an April 2026 "SD4" release. That was wrong. Stability AI's own announcements page lists no SD4 — its 2026 releases are Stable Audio 3.0 (May) and Brand Studio (April); the newest open image family remains SD 3.5. The claim traced to fabricated third-party articles with no primary source. Verdict, models, and the FLUX comparison rewritten to match reality. We verify against primary sources; this one got past us for 20 days and we're leaving the record visible.
- Verified and released from staging. Real change found: Stability shipped SD4 (Base + Ultra) in April 2026 — first full-generation release since SD3.5, moving from U-Net to a diffusion transformer with a dedicated text-rendering module. Updated models, pricing, and the FLUX comparison to account for it. [Retracted 2026-07-29 — see correction above.]
- Added to the stack (self-hosted image seat). Seeded from January 2026 baseline; pricing pending fresh verification by the update loop.
Give Stable Diffusion a job
The self-hosted image seat: open-weight generation with total control, zero marginal cost, and the largest tuning ecosystem.
Try Stable Diffusion →Alternatives with different jobs
Midjourney
03 CreationThe hero-image seat: thumbnails, brand visuals, and any image where aesthetic quality is the deliverable.
Leonardo.Ai
03 CreationThe volume-image seat: daily-token generation, style consistency, and control features for repeatable asset production.
OpenArt
03 CreationThe multi-model image seat: one interface over many image models, with character and style consistency for series work.