# Image One vs Image Anime vs Image Art — which model to pick > The three eroq image models compared on finish, prompt register and settings, with the same subject rendered through all three at the same 10 credits each. Published 2026-09-13 · eroq.ai — canonical: https://eroq.ai/blog/image-one-vs-anime-vs-art eroq has three image models and they all cost the same — 10 credits an image, with negative prompts, CFG control and reference photos on every one of them. That makes the choice unusually clean. There is no budget argument to have, only a finish argument. You pick by what the picture should look like when it lands, and by how you prefer to write. ## Three models, one price - **[Image One](/models/eroq-image-one)** — the photoreal flagship. 1024×1024, 10 credits an image. - **[Image Anime](/models/eroq-image-anime)** — anime, manga and illustration. 10 credits an image. - **[Image Art](/models/eroq-image-art)** — painterly and prose-faithful. It follows long descriptive prompts. 10 credits an image. Everything else is shared. All three take a negative prompt, all three expose CFG, all three accept reference photos to hold a face. Mature imagery of adult, fictional characters is allowed on all three within the acceptable-use policy. In [the Image studio](/studio/image) you get batch 1 to 4, aspect ratios 1:1, 3:4, 4:3, 9:16 and 16:9, and every render lands in the Library with its full recipe so you can Remix it later. ## What each one actually wins at ### Image One — when it has to pass as a photograph Skin, fabric, metal, glass, weather. Anything where the viewer's first question is "is this a photo" and the answer needs to be yes for at least two seconds. Product shots, portraits, location plates, anything that will sit next to real footage in an edit. It is also the safest model when a picture is going to become a video source later. A photoreal still handed to an engine as a start frame gives that engine less to argue with than a painting does. ### Image Anime — when the line is the point Cel shading, manga panels, character sheets, sprite-adjacent illustration, key visuals. The model's job here is not realism but consistency of style: a line weight, a palette, a way of drawing eyes that holds across a set. Use it whenever the brief contains the word "character" in the illustrated sense — a cast for a chat product, a comic, a stream overlay, a set of profile portraits. ### Image Art — when the prompt is a paragraph This is the prose-faithful one, and that phrase is doing real work. Image Art is the model that reads a long descriptive sentence and keeps its clauses. If your prompt has a middle — "…while the first light comes across the field sideways, catching the smoke…" — Art is the model most likely to render the middle rather than average it away. Painterly finish, gallery register, book covers, mood boards, anything where a brushstroke is a feature and not a defect. ## The prompt register — tags versus prose This is the practical difference that trips people up, because the same words do not land the same way in all three. **Image One wants a shot note.** Name a lens, a light and a surface. Write it the way a photographer writes an instruction to an assistant: what is in frame, what the light is doing, what the camera is. Adjectives like "beautiful" or "hyperdetailed" do nothing; "f/1.4", "window light", "worn canvas" do a lot. **Image Anime wants a style plus attributes.** It tolerates a compact comma-separated list far better than the other two — subject, wardrobe, setting, then style terms like "clean line art", "soft cel shading", a palette. You can write full sentences and it will cope, but a tight list of attributes is usually the faster path to the look you had in mind. **Image Art wants the paragraph.** Do not compress. This is the one model where writing more is writing better, because following long descriptive prompts is the thing it was picked for. Give it the scene, the weather, the mood and the painting reference in one flowing piece of prose. A useful rule of thumb: if you find yourself deleting words to make a prompt "cleaner", you are probably writing for Anime. If you find yourself adding a subordinate clause because it matters, you are writing for Art. If you are naming equipment, you are writing for Image One. ## One subject, three models Here is the same idea — a beekeeper at first light — written three ways. Paste any of them straight into the studio. **Image One** > A beekeeper in a worn canvas suit stands between two rows of pale hives at first light, veil pushed back on her forehead, smoke drifting low across the wet grass. Long lens, f/1.4, cold blue shadow with a warm sunrise rim on the veil frame, shallow focus, quiet documentary stillness. **Image Anime** > beekeeper girl, worn canvas suit, veil pushed back, standing between two rows of pale hives, sunrise field, drifting smoke, wet grass, clean line art, soft cel shading, pastel dawn palette, wide establishing composition, calm expression, detailed background **Image Art** > A beekeeper stands very still between two long rows of pale hives while the first light comes across the field sideways, catching the low smoke so that it reads as a warm gray band laid over the blue of the wet grass, her veil pushed back and her hands loose at her sides, painted with a loaded brush in muted dawn colors, the hives softening into suggestion toward the far edge of the field. Three renders, 30 credits, one sitting. The subject survives all three; the finish does not, and that is exactly the point. Run this once with a subject of your own and you will stop guessing which model to open. ## Settings that carry across The controls are the same on all three, but the values you land on differ. - **Negative prompt.** What you are excluding changes per model. On Image One it is usually render artifacts and plastic skin; on Anime it is extra fingers and mangled hands; on Art it is "photograph" creeping back in. See [negative prompt](/glossary/negative-prompt). - **CFG scale.** How hard the model is pushed to obey the prompt. Higher is more literal and more brittle; lower is looser and often prettier. [CFG scale](/glossary/cfg-scale) has the definition. - **References.** Attach a photo to hold a face. Better still, make a [Character](/studio/characters) so the same face is available across images and clips with an @mention. There is more on that in [image generation for characters](/blog/image-generation-for-characters). - **Batch 1–4.** Four variants is 40 credits. On a new subject this is nearly always the right call — one render tells you very little about a prompt. ## Picking in ten seconds - Should it look photographed? **Image One.** - Is it drawn, cel-shaded, or a character in the illustrated sense? **Image Anime.** - Is your prompt a paragraph you do not want compressed? **Image Art.** - Genuinely not sure? Batch the same prompt on two of them. It costs 20 credits to stop arguing. ## FAQ ### Do the three eroq image models cost the same? Yes. Image One, Image Anime and Image Art are all 10 credits per image, and all three support negative prompts, CFG control and reference photos. ### Which eroq image model follows long prompts best? Image Art. It is the prose-faithful model, built to follow long descriptive prompts rather than reduce them to a few keywords. ### Can I use the same reference photo on all three models? Yes. Reference photos work on Image One, Image Anime and Image Art, and a Character's reference photos can be @mentioned in any of them to keep one face consistent across a set. Same subject, three finishes, 30 credits — run the test yourself in [the Image studio](/studio/image).