Two image models now promise magazine-grade stills, and an adult creator's first question is which one will actually draw the scene. MOSS-Image went live on SpicyAPI on 21 October 2026. We ran our content tests on it and on the Krea models, and the results split cleanly.
Short answer: for adult images, pick MOSS-Image. On 21 October it returned all seven of our probe runs as written, from Mild to Maximum, with nothing refused, nothing toned down and no clothing substituted. Krea 2 Large and Medium Turbo refused every prompt from lingerie up to L4 in our September run, Krea 2 Medium softened some and still billed them, and Krea 3 delivered Mild fashion but refused its stronger probes. Krea remains the better fit for clothed editorial looks and wide frames.
Our full MOSS-Image review is at spicyapi.ai/reviews/moss-image.
Two ladders, read side by side
Krea 2 was tested on our September ladder: L1 suggestive, L2 mature themes, L3 strong, L4 intense (an intimate scene between two adults), and L5 extreme (restraint or cinematic gore). MOSS-Image and Krea 3 were tested on the October ladder: Mild, Moderate, Strong, Very strong and Maximum. Both are five-step ladders from suggestive to extreme, but they use different prompt sets and dates. Read the side-by-side as a strong pointer, not a lab comparison.
Adult work: what each model did
| Model | SpicyAPI label | Ladder | What happened |
|---|---|---|---|
| MOSS-Image | Uncensored | October, Mild to Maximum | Seven runs, all delivered as written |
| Krea 2 Large and Krea 2 Medium Turbo | Filtered | September, L1 to L5 | On 24 September all 12 prompts from L1 lingerie to L4 were refused, plus the first L5 probe. The second L5 probe was drawn 3 of 3 |
| Krea 2 Medium | Softened | September, L1 to L5 | Did not refuse. On 16 September its L3 prompts came back in jeans, with underwear added, or with the figure turned away and part of the scene erased, and were billed as normal. Its L1 black lace was delivered on 20 September |
| Krea 3 | Filtered | October | Mild fashion delivered. Both stronger probes, from Moderate up, were refused within about two seconds and not billed |
Two points matter more than the table. First, refused requests are not billed, but softened images are. On Krea 2 Medium you pay for the picture you did not ask for. Second, on 20 September Krea 2 Large and Medium Turbo refused a black lace lingerie shot and a male-subject probe, while a silk slip passed. Even a soft, gallery-style art study at L3 was refused on 16 September.
The Krea 2 Large clothed, low-light register is the one it handles well:
Our full Krea 2 test is in Is Krea 2 uncensored?.
MOSS-Image: what the seven runs covered
The 21 October test used the plain endpoint with no LoRA. It was seven runs in all: one for each of the five levels, a male-subject run at Strong, and a second kind of Maximum probe. Every one came back as written. Each is a single run, so read the result as "it does not refuse", not "every image is perfect". We have not run a blind quality test on MOSS-Image, so we make no ranking claim against Krea.
MOSS-Image is an open-weights model, and the weights do not review prompts. SpicyAPI adds no filter of its own, which is why the label is Uncensored. Any refusal or softening you see comes from the model you chose, not from us.
Lettering: where MOSS-Image is built to help
MOSS-Image is pitched for long posters with readable text, magazine covers with small print, JSON-structured layouts and very long prompts. The poster below came from one prompt.
Prompt tips that help:
- Put each piece of text in quotation marks.
- Say where it goes and how it looks: size, case, colour and typeface style.
- Write line breaks as real new lines.
- Use the length you need. A prompt can run up to 12,000 characters.
We did not run a text-rendering test on Krea, so this section says what MOSS-Image handles. It does not say Krea fails at lettering.
LoRA: MOSS-Image LoRA
MOSS-Image LoRA is the same base model with your own LoRA files added through the loras field. Use it when you need a trained character or style across many images. Without a LoRA loaded, it returned the same picture for the same prompt and seed as plain MOSS-Image, so a LoRA changes the output only once you load one. The LoRA page is at MOSS-Image LoRA.
Prices and sizes
At the time of writing (22 October 2026), these are the per-image prices. The model page always shows the current price.
| Model | Price per image | Notes |
|---|---|---|
| MOSS-Image | $0.024 at 1k, $0.048 at 1.5k, $0.096 at 2k | 15 aspect ratios, default 1:1; JPEG, PNG or WebP; seed |
| MOSS-Image LoRA | $0.03 at 1k, $0.054 at 1.5k, $0.102 at 2k | Your LoRA files through loras |
| Krea 2 Medium Turbo | $0.015 | 2.35:1 frames; up to 10 style references, plus $0.005 per request; no seed |
| Krea 2 Medium | $0.03 | No seed |
| Krea 2 Large | $0.06 | 2.35:1 frames; up to 10 style references, plus $0.005 per request; no seed |
| Krea 3 | $0.036 text-to-image | Edit: $0.036 per source image, $0.072 with one source |
MOSS-Image's 1k sizes are 1024x1024 at 1:1 and 1360x768 at 16:9. Its 1.5k 16:9 size is 2048x1152, and its 2k 9:16 size is 1536x2720.
At 1k, MOSS-Image costs $0.024 an image. Krea 2 Medium Turbo is the lowest per-image price at $0.015, but it refused the September adult prompts, so a low price does not help a shot that never arrives.
Krea 3: clothed looks, unstable edits
Krea 3 went live on 15 October 2026 with a Filtered label. It delivered Mild fashion. It is a good fit for clothed fashion, fabric and skin texture, and period film looks. Its editing boundary is unstable: in our 16 October tests the same edit passed once and was refused on a retry with another seed.
Editing: MOSS-Image cannot
MOSS-Image is text-to-image only. It has no editing, no reference images, no negative prompt, no guidance or step controls, and one image per request. If you need to change an existing picture, look at Is Seedream uncensored? instead. Krea 3 does edit, with the unstable boundary above.
Which one should you pick?
- Adult images from a text prompt: MOSS-Image.
- A trained character or style: MOSS-Image LoRA, from $0.03 at 1k.
- Posters, magazine covers and quoted lettering: MOSS-Image.
- Clothed editorial looks, wide 2.35:1 frames and style references: Krea 2 Large or Krea 2 Medium Turbo.
- Clothed fashion, fabric and skin texture, period film looks: Krea 3 at $0.036.
- Editing an existing image: not MOSS-Image. Use a model with an edit task.
Questions people ask
Is MOSS-Image uncensored? SpicyAPI labels it Uncensored. In our 21 October runs all seven came back as written, from Mild to Maximum. That is seven runs, not a guarantee for every prompt.
Is MOSS-Image better than Krea 2? For adult work in our tests, yes on refusals: MOSS-Image delivered every level we ran, while Krea 2 refused or softened many of ours. We make no claim about overall image quality, since MOSS-Image has not had a blind quality test.
Can MOSS-Image edit photos? No. It is text-to-image only. Use a model that has an edit task.
Can I use my own LoRA with MOSS-Image? Yes, through MOSS-Image LoRA. Your LoRA files go in the loras field, at $0.03 at 1k, $0.054 at 1.5k and $0.102 at 2k.
Adults only: use adults in every prompt and upload, do not generate a real person's likeness without their consent, and you are responsible for what you generate.
