Run the actual checkpoints.
4 credits · free tier includes 100 · a failed run costs nothing
Most sites with “Stable Diffusion” in the name quietly serve you something else. This one runs real community checkpoints, addressed by their Civitai id — the same weights you’d download yourself, minus the 12 GB, the Python environment, and the GPU. Verified and live today, and open to any checkpoint you ask for.
The same instruction, 9 checkpoints.
This is the comparison people keep a local install for, and the reason a named checkpoint list matters more than a faster button. Same prompt, same catalogue, nine different ideas of what it means.
anime portrait of a knight in ornate armor, dramatic lighting, highly detailed, illustration
2 SDXL · 4 Pony · 3 SD 1.5 — FLUX sits this one out: its single checkpoint drew a different prompt, and captioning it as the same one is exactly the thing this page is about.
Everything that is live right now.
86 images below, every one generated by the checkpoint named under it — no stock photography and no cover art borrowed from a different model. Download counts are Civitai’s own.
The original lineage. Fast, cheap, native around 512² — and still the deepest catalogue of fine-tunes anywhere.
The workhorse. Native 1024², the widest LoRA pool, and where most photoreal work lives today.
An SDXL fine-tune with its own prompt grammar — score tags up front. Same tensor shape, so it shares SDXL’s LoRAs.
A different architecture entirely. Strong prompt adherence, no negative prompt needed — and its own separate LoRA pool.
| Checkpoint | Family | Best at | Civitai downloads | Steps |
|---|---|---|---|---|
| Realistic Vision V6.0 civitai:4201@130072 | SD 1.5 | Top SD1.5 photoreal model | 2,236,166 | 28 |
| DreamShaper 8 civitai:4384@128713 | SD 1.5 | Versatile SD1.5 classic | 1,624,820 | 28 |
| Juggernaut XL civitai:133005@1759168 | SDXL | Most-downloaded SDXL. Crisp photoreal | 1,521,404 | 30 |
| majicMIX realistic civitai:43331@176425 | SD 1.5 | Asian photoreal portraits | 1,219,904 | 28 |
| Pony Diffusion V6 (Turbo) civitai:257749@298112 | Pony | The Pony base, fast turbo merge. Stylized characters (use score_9 tags) | 990,423 | 10 |
| epiCRealism civitai:25694@143906 | SD 1.5 | Natural-light photoreal | 865,447 | 28 |
| RealVisXL V5.0 civitai:139562@154590 | SDXL | Photoreal portraits + scenes | 631,066 | 30 |
| DreamShaper XL civitai:112902@121931 | SDXL | Versatile illustrative + stylized output | 608,769 | 30 |
| ReV Animated civitai:7371@425083 | SD 1.5 | Fantasy / semi-anime illustration | 574,133 | 28 |
| Pony Realism civitai:372465@914390 | Pony | Photoreal Pony fine-tune | 544,630 | 28 |
| MeinaMix civitai:7240@119057 | SD 1.5 | Clean anime SD1.5 | 524,505 | 28 |
| Beautiful Realistic Asians civitai:25494@177164 | SD 1.5 | SD1.5 Asian photoreal portraits | 503,797 | 28 |
Showing the 12 most-downloaded of 32
If it has a Civitai id, it can run here.
32 checkpoints are live and verified today. That is not the limit — it is just what has been through the check so far. Models are addressed by their Civitai id, so what can run here is bounded by the open catalogue, not by a list somebody has to maintain.
Paste the model page or the raw civitai:1188071@1408658 form. Requested checkpoints are usually live the next day.
- 01You give the id
A Civitai link or the raw model@version id. That string is what we send — nothing gets re-hosted, re-trained, or substituted.
- 02It gets resolved
We check the provider has that exact version indexed and that it is a base checkpoint rather than an inpainting or refiner build.
- 03It gets generated against
One real image, on the actual weights. An id that resolves is not the same thing as a model that loads, and this is the step that catches the difference.
- 04It goes live
With its own page, its own defaults for its architecture, and its id printed where you can check it.
Checkpoints the provider has not indexed. The whole Illustrious family is in this bucket right now — we re-check, and it is not listed on this site until it actually runs. A model that cannot be generated against does not get a page here, however much traffic its name would bring.
Adult models. A large and profitable part of this ecosystem, deliberately left out. That is a decision, not an oversight, and it is not going to change.
What “Stable Diffusion online” usually means now.
The term has drifted. Most sites ranking for it today route your prompt to a general-purpose commercial model and never say which one. That is a fine product — it is just not the thing you searched for.
The common setup
- 01One unlabelled “Base” model behind the box.
- 02The model list is whatever they decided to ship. No way to ask.
- 03No checkpoint choice, no LoRA, no negative prompt.
- 04Credit cost per generation shown after you spend it.
- 05A failed or rejected run still burns your quota.
Here
- 0132 named checkpoints, each with its Civitai id printed on the page.
- 02Any checkpoint with a Civitai id can be added — usually the next day.
- 03Checkpoint, style, steps, size and negative prompt — all yours.
- 04Credit cost shown on the button before you press it.
- 05A run that fails or gets rejected is refunded automatically.
One credit per image. That is the entire price list.
Not one price for the cheap models and another for the good ones, and no separate table for the checkpoint you actually wanted. Every checkpoint here runs at the same rate, and the cost is printed on the button before you press it.
- All 32 checkpoints and all 11 styles — nothing is held back for a paid tier
- No watermark, and we claim no rights over what you generate
- Your images stay private unless you publish them
The first image does not ask you to sign up. Type a prompt, pick a checkpoint, generate — the account only comes up when you want to keep the result, change models, or go past the free allowance.
A run that fails or gets rejected is refunded automatically. You spend credits on results you keep, and on nothing else.
Generate your first imageStable Diffusion online, answered straight.
The things worth knowing before you generate anything — including the two this site cannot do.
- Is this actually Stable Diffusion, or a different model behind the name?
- It is actually Stable Diffusion. Every checkpoint here is a community model from the open Civitai catalogue, addressed by its Civitai id — the same identifier is printed on the page next to the model name, so you can look up the exact build we run. The catalogue covers the four base families in real use today: SD 1.5, SDXL, Pony (an SDXL fine-tune) and FLUX.1 D. This is worth checking wherever you generate: several sites that rank for “Stable Diffusion online” route your prompt to a general-purpose commercial model instead and do not say which one.
- Do I need to install anything, or own a GPU?
- No. Everything runs in the browser — there is no 12 GB checkpoint download, no Python environment, no CUDA setup, and no graphics card requirement. The same weights you would otherwise download from Civitai and load into a local WebUI or ComfyUI install are executed on hosted GPUs instead.
- Which Stable Diffusion models can I run?
- 32 checkpoints are verified and live right now, including Juggernaut XL, RealVisXL, DreamShaper, Realistic Vision, majicMIX realistic, epiCRealism, Animagine XL, Pony Diffusion V6 and PixelWave. That is not a fixed list: models are addressed by their Civitai id, so any checkpoint in the open catalogue can be added — usually within a day of being requested.
- What is the difference between SD 1.5, SDXL, Pony and FLUX?
- They are different base architectures, and the choice has practical consequences. SD 1.5 is the original lineage — fast, cheap, native around 512×512, with the deepest catalogue of fine-tunes. SDXL is native around 1024×1024 and is where most photoreal work lives. Pony is an SDXL fine-tune with its own prompt grammar (score tags at the front) that shares SDXL’s tensor shape. FLUX.1 D is a separate architecture with strong prompt adherence that needs no negative prompt. The tensor family also decides which LoRAs will load at all — a LoRA only attaches to a base of its own family.
- Can I use LoRAs and negative prompts?
- Yes. Negative prompts are available on every SD 1.5, SDXL and Pony checkpoint, pre-filled with a sensible default for that architecture, and FLUX runs without one because it does not benefit from it. 11 style LoRAs are available today, all SDXL-tensor, so they attach to any SDXL or Pony base. Steps, output size and seed are all shown as well, rather than hidden behind a quality preset.
- Is it free, and what does a generation cost?
- The free tier includes 100 credits and does not ask for a card. Every checkpoint costs one credit per image — there is no separate rate for the better models, and the cost is printed on the generate button before you press it rather than deducted afterwards. A run that fails or gets rejected is refunded automatically.
- Who owns the images, and can I use them commercially?
- We add no watermark and claim no rights over what you generate, and nothing you make is published anywhere unless you choose to publish it. Note that each community checkpoint carries its own licence on Civitai, and a few of them do restrict commercial use — that licence is set by the model’s author, not by us, so check the model’s Civitai page if the work is commercial.
- Can I run Illustrious or NoobAI checkpoints?
- Not yet. The Illustrious family is not currently indexed by the provider that executes these models, so it cannot be run here — and a checkpoint that cannot actually generate does not get listed on this site, however much traffic its name would attract. It is re-checked periodically.
- Is adult content allowed?
- No. Adult models and adult generations are out of scope here. That is a deliberate decision rather than an oversight, and it is not going to change — a large and profitable part of the Civitai ecosystem is simply not served by this site.