Skip to navigation

Choose a Model

Every v2 endpoint names its model in the path, POST /v2/{content}/{action}/{model}, so choosing a model means choosing an endpoint. Start from the task, then pick the model that has the inputs and controls you need.

Pick an endpoint for the task

TaskEndpointWhat to know
Generate images from a promptPOST /v2/image/generate/{model}Compare the models below, or send the request to auto and let Ideogram pick
Edit an image and keep its sizePOST /v2/image/precise-edit/ideogram-4-5The result has the source image’s width and height. Optional mask and up to 4 reference images
Transform an image with a promptPOST /v2/image/remix/{model}image_weight, from 1 to 100, sets how closely the result follows the source
Repaint a masked regionPOST /v2/image/inpaint/ideogram-3The mask is required. Black marks the area to repaint
Change the aspect ratioPOST /v2/image/reframe/{model}Ideogram 3.0 keeps the source pixels in the center and fills the new area
Replace the backgroundPOST /v2/image/replace-background/{model}Ideogram 3.0 and GPT Image 2 follow a prompt. Ideogram 4.5 copies the setting in a background_reference image
Remove the backgroundPOST /v2/image/remove-background/ideogram-1Returns the foreground as a transparent PNG
Remove an objectPOST /v2/image/remove-object/ideogram-1The mask is required. White marks the object to remove. Returns a generation_id to poll
UpscalePOST /v2/image/upscale/{model}2x, 4x or 8x. See upscale models
Turn an image into a promptPOST /v2/image/describe/{model}Ideogram 4.0 returns a structured JSON prompt; Ideogram 3.0 returns text
Make the text in an image editablePOST /v2/design/layerize/ideogram-3Returns positioned text blocks with fonts, sizes and colors, plus the image without its text
Generate SVG artworkPOST /v2/svg/generate/{model}Ideogram Vector 1, or GPT Image 2.6 Flare and Sunburst
Convert an image to SVGPOST /v2/svg/vectorize/ideogram-vector-1Runs asynchronously
Generate videoPOST /v2/video/generate/{model}Always asynchronous. See Generate video
Adapt ads and product photosPOST /v2/tool/{tool}Resize, localize and vary ads; recolor and re-material products. Always asynchronous

To edit or generate in the Ideogram app instead, see Models in the user guide.

Image generation models

ModelEndpointChoose it forInputs and controls
AutomaticautoNot choosing a model yourselfIdeogram picks a model per request. Up to 10 source images to edit, and a resolution_tier of 1k, 2k or 4k
Ideogram 4.6ideogram-4-6The newest Ideogram model, for generating or editing at up to 4KText or JSON prompts, up to 5 source images and a mask, background (auto, transparent, opaque), quality (very_low, low, medium, high) and exact sizes up to 4096x4096
Ideogram 4.5ideogram-4-5Editing or combining images with a prompt, and transparent output from a promptText or JSON prompts, up to 5 source images and a mask, background (auto, transparent, opaque) and quality (very_low with source images, low, medium, high)
Ideogram 4.0ideogram-4Wide formats such as banners and headers, at 1K or 2KText or JSON prompts, rendering_speed (turbo, default, quality)
Ideogram 4.0 Transparentideogram-4-transparentStickers, icons and print designs with no backgroundoutput_resolution from 1k to 8k
Ideogram 3.0ideogram-3Logos, brand graphics and matching a set styleStyle types, presets and codes, up to 10 style_reference_images, a color palette and a negative prompt
Ideogram 3.0 Characterideogram-3-characterKeeping one character consistent across imagesOne character_reference_images image and an optional mask
Ideogram 3.0 Transparentideogram-3-transparentTransparent output with Ideogram 3.0 stylingupscale_factor of 1x, 2x or 4x in the same call
Custom modelsideogram-4-custom-model, ideogram-3-custom-modelA look trained on your own imagescustom_model_uri. Training uses the v1 API
P-Image Ideogramp-image-ideogramFast, low-cost drafts and volumeQuality from very low to very high, 1k or 2k, or custom dimensions
GPT Imagegpt-image-2, gpt-image-2-5-flare, gpt-image-2-5-sunburstOpenAI’s models, editing up to 16 source imagesA mask on the first source image limits the edit. Flare is the fast variant of GPT Image 2.5 and Sunburst the quality variant; both add background and quality
Nano Banananano-banana-2, nano-banana-2-1, nano-banana-proGoogle’s models, with up to 10 source imagesresolution_tier of 1K, 2K or 4K
Z-Imagez-imageExact output sizes up to 1,048,576 pixelsnum_inference_steps from 1 to 128
Ideogram 2.0 and 2aideogram-2, ideogram-2aExisting integrationsStill available. Use Ideogram 4.5 or 4.0 for new work

Every generate endpoint takes num_images from 1 to 8. Ideogram models also take seed, which makes a generation repeatable, and most take magic_prompt, which rewrites your prompt with more detail before the model sees it. Turn it off when the model should follow your exact words.

What image generation costs

Each model has its own rate, and some rates depend on quality, speed or output size. These are the rates per image on the API pricing page on November 15, 2026. The pricing page is the current source, and a price quote gives the exact price of one request.

ModelPrice per image
Ideogram 4.6, no source images$0.008 very low, $0.03 low, $0.06 medium, $0.10 high up to 2K; $0.06 low, $0.12 medium, $0.20 high at 4K
Ideogram 4.5, no source images$0.03 low, $0.06 medium, $0.10 high
Ideogram 4.0$0.03 turbo, $0.06 default, $0.10 quality
Ideogram 3.0$0.03 turbo, $0.06 default, $0.09 quality
Ideogram 3.0 Character$0.10 turbo, $0.15 default, $0.20 quality
P-Image Ideogram, 1K$0.003 very low to $0.033 very high
Nano Banana 2$0.0672 at 1K, $0.1008 at 2K, $0.1512 at 4K
Nano Banana Pro$0.1344 at 1K or 2K, $0.24 at 4K
GPT ImageOpenAI’s list token prices for the tokens each request uses

Ideogram 4.6 without quality renders at medium. Ideogram 4.5 without quality renders at high when you send no source images, and at medium when you do.

Upscale models

ModelEndpointChoose it for
AutomaticautoLetting Ideogram pick per request. The choice may change over time
Topaz Standard V2topaz-standard-2A clean enlargement with no prompt
Topaz Text Refinetopaz-text-refineKeeping lettering sharp on ads, labels and packaging
Topaz Wonder 3.5topaz-wonder-3-5Rebuilding detail, with enhancement_strength
Topaz Wonder 4topaz-wonder-4Removing noise, blur and compression artifacts and recovering faces in one pass, at 2x or 4x
Topaz Bloom 2 and Topaz Redefinetopaz-bloom-2, topaz-redefineCreative upscaling that regenerates detail, guided by an optional prompt
Nano Banana Pronano-banana-proRegenerating the image at a 1K, 2K or 4K tier

Topaz upscalers bill per output megapixel. The auto, Standard V2, Text Refine, Wonder 3.5, Bloom 2 and Redefine endpoints reject a request whose output would be wider or taller than 8,192 pixels. The user guide’s Upscale page compares faithful and creative upscalers with examples.

Video models

ModelEndpointsInputsLength
Seedance 2.5Text to video, image to video, reference to videoFirst and last frames, up to 9 reference images and up to 10 reference audio clips4 to 30 seconds
Seedance 2.0Text to video, image to video, reference to videoFirst and last frames, up to 9 reference images and 3 reference audio clips4 to 15 seconds
Kling 4.0One endpointUp to 10 frames and 10 reference images3 to 30 seconds
Kling v3 StandardText to video, image to videoFirst and last frames3 to 15 seconds
MiniMax H3Text to video, image to video, reference to videoFirst and last frames, up to 9 reference images and 3 reference audio clips5 to 15 seconds

Generate video covers the request, polling and pricing.

Vector models

ModelEndpointChoose it for
Ideogram Vector 1POST /v2/svg/generate/ideogram-vector-1SVG artwork from a prompt and up to 5 image references, at low, medium or high quality
Ideogram Vector 1POST /v2/svg/vectorize/ideogram-vector-1Converting a raster image to SVG, in simple, detailed or auto mode
GPT Image 2.6 Flare and SunburstPOST /v2/svg/generate/gpt-image-2-6-flare, POST /v2/svg/generate/gpt-image-2-6-sunburstSVG artwork from OpenAI’s models, with quality and thinking_level

Older models

The v1 API and Ideogram 2.0 and 2a keep working. See Model lifecycle and deprecations for what is legacy and how deprecations are announced.