AI image model · Ideogram
Ideogram 4.0, Explained: Open Weights, License, Price
Ideogram 4.0 is Ideogram's first open-weight text-to-image model, released on June 3, 2026; this page covers its specs, license and pricing. Need product photos and edits instead of a model to host? Supavisual's own generator costs 1 credit per image.
- Supavisual: 3 free trial credits
- Commercial license on paid plans
- No GPU or self-hosting needed
- 9.3B
- Parameters in the open Ideogram 4.0 weights
- 2,048 px
- Largest native side, any multiple of 16
- $0.03
- Cheapest API image (Turbo) as of November 2026
- 1 credit
- Per image on Supavisual's own generator
What is Ideogram 4.0?
Ideogram 4.0 is a text-to-image model from Ideogram, the company known for rendering readable text inside images. Released on June 3, 2026, it is the company's first model with downloadable weights, and Ideogram says it was trained from scratch rather than fine-tuned from another checkpoint.
The design goal is controllable, design-grade output. The model was trained only on long, structured JSON captions that describe every object, text region and style choice in an image. At inference you mirror that format: a high-level description, a compositional breakdown, optional bounding boxes that pin each element to a position, and a hex color palette. Ideogram 4 also renders multilingual text and outputs native 2K images.
One naming note: Ideogram has since listed Ideogram 4.5 as a newer model. Version 4.0 remains the open-weight release, which is why it is the one developers download, fine-tune and build on.
Ideogram 4.0 specs and architecture
Figures from Ideogram's own model documentation. The last row shows Supavisual's generator for comparison.
| Maker | Ideogram |
|---|---|
| Released | June 3, 2026 |
| Parameters | 9.3B |
| Architecture | Flow-matching, fully single-stream diffusion transformer with 34 layers |
| Text encoder | Qwen3-VL-8B-Instruct vision-language model, features from 13 layers |
| Resolution | 256 to 2,048 px per side in multiples of 16, aspect ratios up to 6:1 |
| Prompt format | Structured JSON with optional bounding boxes and a hex color palette |
| Public weights | nf4 (CUDA, Diffusers support) and fp8 (all hardware, no Diffusers) |
| Alternative: Supavisual's generator | Hosted, prompt plus up to 6 references, 1–4 images, 1 credit each |
Where Ideogram 4.0 is strongest
Ideogram built the model for brand and layout work, and its strengths follow from that.
Text in the image
Signage, logos, captions and multi-line text are the core strength. Ideogram says it leads every open-weight model it benchmarked on text, including larger ones such as Qwen-Image (20B) and FLUX.2 [dev] (32B).
Bounding-box layout
Put coordinates in the prompt to fix where the subject, headline and background sit. Dense posters keep their structure instead of drifting.
Color palette control
List hex colors in the style description and the image's dominant scheme follows them, which helps keep brand colors on target.
Odd aspect ratios
Any size from 256 to 2,048 px per side and ratios up to 6:1 cover phone wallpapers, ultrawide banners and square posts with one model.
Fine-tuning
Because the weights are public, you can fine-tune on your own style guide or product shots; fine-tunes are permitted under the non-commercial agreement.
Runs on your hardware
Self-host behind your own firewall. Inference cost then follows the GPUs you provision rather than a per-image fee.
Is Ideogram 4.0 open source? The license explained
The weights are open to download, but commercial use is licensed separately. Ideogram offers four routes.
| Route | Commercial use | Self-host | Volume |
|---|---|---|---|
| Open weights (Non-Commercial Model Agreement) | Not permitted | Yes, non-commercial only | Research, evaluation, personal |
| Self-serve commercial license | Included | Yes, public quantized weights | 10,000 to 100,000 images a month |
| Enterprise license | Custom terms | Yes, full precision available | 100,000+ images a month |
| Ideogram API | Included | No, hosted | Pay per image |
Terms as of November 2026. Outputs from the free weights are yours, subject to Ideogram's acceptable use policy; read the agreement for your exact use case.
Ideogram 4.0 pricing: API and app (as of November 2026)
The hosted API charges per image by rendering tier. The Ideogram app uses credits instead.
| Tier | API, per image | App credits, per 2K image |
|---|---|---|
| Turbo | $0.03 | 3 credits |
| Default / Balanced | $0.06 | 6 credits |
| Quality | $0.10 | 10 credits |
| Transparent, 4K | $0.19 to $0.26 | Not listed |
| Transparent, 8K | $0.35 to $0.42 | Not listed |
App plans as of November 2026: pay as you go at 1 cent per credit, Plus $20 a month (2,400 priority credits), Pro $60 a month (7,500), Team $30 per user (3,600 each); yearly billing is lower. Free accounts get weekly credits when signing in with Google, Apple or Microsoft.
How to run Ideogram 4.0: app, API or your own GPU
Pick the route by what you need to own. The Ideogram app is the fastest test: choose 4.0, pick Turbo, Balanced or Quality and spend 3, 6 or 10 credits per image. The Ideogram 4.0 API gives hosted commercial access at $0.03 to $0.10 per image with no subscription, which is the simplest way to put it inside a product.
To self-host, accept the license gate on the Hugging Face model page, download the nf4 or fp8 weights and run Ideogram's inference code. For the highest quality, Ideogram suggests 2048 × 2048 with its V4_QUALITY_48 sampler preset. The model expects JSON prompts, so most setups use a language model to turn a plain request into the caption schema.
Remember the license line: free weights are for research, evaluation and personal projects. Ideogram's licensing table marks production and commercial use of them as not permitted, so a commercial self-hosted deployment needs the self-serve or enterprise license.
Ideogram 4.0 weaknesses and common mistakes
Most disappointing results come from treating it like a plain-prompt model. Avoid these.
Do
- Write prompts in the JSON caption schema, or convert them with an LLM
- Name every object you want; the model renders what the caption lists
- Use bounding boxes for headlines that must land in one spot
- Budget for a commercial license before you ship self-hosted output
Don't
- Expect a short one-line prompt to match JSON results
- Use the free weights for client or paid work
- Assume full-precision weights; the public files are quantized
- Wait on native alpha layers; Ideogram lists them for a later 4.0 release
An Ideogram 4.0 alternative without self-hosting
Supavisual does not run Ideogram's model. It is a separate hosted tool with its own generator, for teams that want finished product and marketing images without GPUs, JSON schemas or a second license. In the generator you write a normal sentence, add up to 6 reference images, choose a ratio and get up to 4 images, at 1 credit each.
For text-heavy work, the infographics tool lays out graphics, and translate rewrites the text already in an image into 37 languages. The editor removes backgrounds for transparent cutouts, upscales and expands, and batch applies one edit to up to 10,000 images.
Hobbyist is $19 a month for 800 credits, about 2.4 cents per image, and $9 the first month. Unlimited is $99 per seat per month. Both include a commercial license, so there is no extra licensing step. New accounts get 3 trial credits with watermarked previews.
Ideogram 4.0 vs GPT Image 2.6, Nano Banana 2.1 and Seedream 5.5
The only open-weight model in this group, next to three hosted ones.
| Ideogram 4.0 | GPT Image 2.6 | Nano Banana 2.1 | Seedream 5.5 | |
|---|---|---|---|---|
| Made by | Ideogram | OpenAI | ByteDance | |
| Weights | Open (non-commercial) | Hosted only | Hosted only | Hosted only |
| Stand-out feature | Text, layout boxes, hex palettes | SVG and GIF output | 14 references, 4K | Pixel-level multi-turn edits |
| Published API price | $0.03–$0.10 per image | $30 / 1M output tokens | $0.0336 per 1K image | Not listed by ByteDance |
Prices as of November 2026. Read more: GPT Image 2.6, Nano Banana 2.1, Seedream 5.5.
More models and Supavisual tools
Other model explainers and the Supavisual tools for finished marketing visuals.
- GPT Image 2.6OpenAI's November 2026 image model: Sunburst and Flare, GIF and SVG output, API pricing and limits.
- Nano Banana 2.1Google's October 2026 Flash image model: specs, per-image API prices, limits and what Reddit thinks.
- Seedream 5.5ByteDance Seed's newest image model: multi-turn editing, transparent layers, lifelike texture and where to try it.
- ArtbreederSupavisual vs Artbreeder: features, pricing as of November 2026, and which one suits product and marketing images.
- GenerateTurn a prompt into a finished product or marketing image, 1 credit each.
- InfographicsTurn product details into an infographic image.
- PricingHobbyist and Unlimited plans, with 3 free trial credits to start.
Frequently asked questions
What is Ideogram 4.0?
It is Ideogram's first open-weight text-to-image model, released on June 3, 2026. The 9.3B-parameter model was trained from scratch on structured JSON captions, which gives it strong in-image text, bounding-box layout control, hex color palettes and native output up to 2,048 px per side. It is available in the Ideogram app, through the API and as downloadable weights.
Is Ideogram 4.0 free?
The weights are free to download under Ideogram's Non-Commercial Model Agreement, for research, evaluation and personal projects. Commercial use needs the API, a self-serve license or an enterprise deal. In the app, free accounts receive weekly credits when signing in with Google, Apple or Microsoft, and a 2K Turbo image costs 3 credits.
Can I use Ideogram 4 commercially?
Yes, but not with the free weights alone. The hosted API includes commercial use and charges $0.03 to $0.10 per image. To self-host commercially you buy a self-serve license sized at 10,000 to 100,000 images a month, or an enterprise license above that, which can also include full-precision weights. Terms as of November 2026.
How much does the Ideogram 4.0 API cost?
As of November 2026 the API charges $0.03 per Turbo image, $0.06 per Default image and $0.10 per Quality image, with no subscription. Transparent output costs the same at 1K and 2K, and rises to $0.19 to $0.26 at 4K and $0.35 to $0.42 at 8K, depending on the tier you choose.
What hardware do the Ideogram 4.0 weights run on?
Ideogram publishes two quantized versions. The nf4 weights run on CUDA GPUs and work with Diffusers; the fp8 weights run on all supported hardware but without Diffusers support. Both are gated on Hugging Face behind a license agreement. Ideogram does not list a minimum memory figure on its model page, so test on your own GPU.
Why does Ideogram 4.0 need JSON prompts?
Because it was trained only on structured JSON captions that name every object, text region and style choice. Mirroring that schema is the most reliable way to get each requested element rendered, and it unlocks bounding boxes and color palettes. Most people let a language model convert a plain-English request into the JSON format.
Does Supavisual use Ideogram 4.0?
No. Supavisual is an independent hosted tool with its own generator and does not offer Ideogram's model. You prompt in plain English, add references, then remove backgrounds, upscale, resize or translate on-image text at 1 credit per image. Hobbyist is $19 a month for 800 credits, and paid plans include a commercial license.
Try an Ideogram 4.0 alternative with no setup
Read up on Ideogram's open model, then see what Supavisual makes from a plain-English prompt or product photo: 1 credit per image, 3 free trial credits.
Generate an imageMore ai image models: see all.