Ideogram 4.6 adds native 4K images, with high-quality generation at $0.20
Edited and reviewed by AI Ping Editorial Team
3-line summary
- The November 19 release generates native 4K images without an upscaling step.
- High-quality 4K API generation costs $0.20 per image, versus $0.60 for editing.
- Regional edits can preserve a large original’s dimensions, subject to an edit-area limit.
Ideogram released its 4.6 image model on November 19 in the app and API. The company says it generates native 4K images without upscaling a smaller picture.
The official model page prices high-quality 4K API generation at $0.20 per image and editing at $0.60. Very low quality starts at $0.008 up to 2K, without 4K support. Ten high-quality 4K outputs therefore cost $2 for generation or $6 for editing.
The app’s credit table lists high-quality 4K generation at 20 credits per image, or 60 credits when source images are supplied. Subscription credits and API dollar rates describe their respective billing routes. Ideogram lists subscriptions, on-demand credits and hosted API usage separately.
References and edit regions have different roles
The Generate API accepts a text prompt alone or up to five attached images. With images, it modifies the first and uses the others as references. Product photographs can be supplied with color or material references. Both natural-language descriptions and structured JSON prompts are accepted. Applications can supply an image description in a machine-readable form, alongside the photographic references.
Generation offers automatic, transparent or opaque backgrounds; transparency has an alpha channel. Sizes can be automatic, 2K, 4K or exact dimensions. Automatic defaults to 2K without source images.
Ideogram’s model guide distinguishes selecting 4.6 from its automatic model option, which chooses a model for each request. The 4.6 API defaults to medium quality, with high quality selected separately. The four levels are very low, low, medium and high. High renders several candidates and selects the best, taking longer than the single render at medium. Requests can ask for one to eight outputs. Billing is per generated image, so a batch changes the number of charged outputs.
Whole-image Precise Edit has a 16,777,216-pixel limit, the area of a 4096-square. An entire 8192-square source is scaled down proportionally. Very low quality reduces the limit to a 2048-square, with no side above 4608 pixels.
Selecting a region by mask or coordinates preserves large source dimensions and leaves its surroundings unchanged. A same-size mask marks edits in black and preserved areas in white. Masked requests allow up to three additional references. Billing follows source size; selecting a smaller region does not lower its rate.
Rapidata’s multi-step image editing leaderboard lists 4.6 at 1,892 and 4.5 at 1,866. Their reported 95% intervals, ±32 and ±30 respectively, overlap. Settings differ: 4.6 uses high-quality Precise Edit, while 4.5 uses high quality and high precision.
The evaluation applies five consecutive edits to 199 photographs of real people, changing clothing, objects or text rather than faces. Human judges compare faces after the final edit with the originals. The ranking measures identity preservation. Its settings table records 1248-by-832 input and output for both Ideogram versions, rather than native 4K evaluation.
In the app, users select 4.6 when available and receive a request-specific credit quote. AI editing and image references require priority credits. Unused subscription credits expire monthly; top-ups roll over. The slow queue after priority credits run out excludes those editing features.
Sources
- Ideogram 4.6 — release, native 4K and API prices — Ideogram
- Generate with Ideogram 4.6 — images, output and sizes — Ideogram
- Precise Edit with Ideogram 4.6 — region and source-size limits — Ideogram
- Model selection and default quality — Ideogram
- Plans, credits and API billing — Ideogram
- Multi-step Image Editing — leaderboard — Rapidata
- Multi-step Image Editing — methodology — Rapidata
- Multi-step Image Editing — model settings — Rapidata
- Official model distribution: inputs, quality and output behavior — Ideogram on Replicate