Ideogram 4.0 arrived on June 3 with a claim that sounded unlikely a month ago: an open-weight image model that matches the quality of GPT Image 2 and Gemini’s Nano Banana 2. It pulled it off. The model is a 9.3 billion parameter foundation model you can download, fine-tune, and run on your own hardware. And it does something the closed models still will not: give you layout control, colour conditioning, and production-grade text rendering as native features.<\/p>\n
Three capabilities set this apart from every other open-weight image model available today.<\/p>\n
Text rendering that holds up in real design work.<\/strong> Ideogram has led on typography since its first release. Version 4.0 extends that with multilingual support, denser type at smaller scales, and reliable rendering of headlines, packaging copy, and signage. In a blind evaluation by ten professional designers from Contra’s top-earning talent, Ideogram 4.0 was picked as the best model 47.9% of the time, well ahead of Gemini Nano Banana 2 at 30.0%, FLUX.2 max at 15.5%, and Grok Imagine 1.0 at 15.0%. When the same designers were asked whether they would use the output in real client work, Ideogram 4.0 scored 3.55 out of 5. Nano Banana 2 scored 2.84.<\/p>\n Bounding-box layout control.<\/strong> You tell the model where a logo, headline, or subject belongs on the canvas using normalised coordinates. Layout is no longer something the model guesses and you correct afterwards. You direct it. On the 7Bench layout benchmark, Ideogram 4.0 is significantly better than every closed-source model tested.<\/p>\n Colour palette conditioning.<\/strong> You can specify up to 16 hex colours per image, and up to 5 per element, to steer the dominant colour scheme directly. No more writing “warm muted earth tones” and hoping the model interprets it the same way you do. You give it #8B4513 and it uses #8B4513.<\/p>\n Most image models treat the prompt as a sentence. Ideogram 4.0 treats it as a structured object. The model was trained exclusively on structured JSON captions, and the inference pipeline parses every prompt as JSON before generation.<\/p>\n Each JSON prompt contains a style block, optional bounding boxes, a colour palette, and typed text elements. Each text element carries the literal string to render plus a separate visual description for its styling. This is what makes clean multi-line, multi-font text in images actually work.<\/p>\n If writing JSON by hand sounds tedious, the inference pipeline ships with a “magic prompt” step. It uses an LLM to expand a casual sentence into a full structured caption before generation. By default it calls Ideogram’s hosted API, but the system prompt is open source so you can run the expansion through your own provider.<\/p>\n A single set of weights covers any resolution from 256 to 2048 pixels in multiples of 16, with aspect ratios up to 6:1. The noise schedule auto-adjusts per resolution. For the highest quality output, set height and width to 2048 and use the V4_QUALITY_48 sampler preset, which runs 45 steps at guidance weight 7 followed by 3 polish steps at guidance weight 3. Faster presets exist for iteration: V4_DEFAULT_20 runs 20 steps with two polish steps, and V4_TURBO_12 runs 12 steps with a single polish step.<\/p>\n This is the feature that could change how design work actually ships. Ideogram 4.0 is the start of a generation stack built around the reality that professional design work is not a single pixel layer. Today, the Background Remover produces a clean alpha cutout from any 4.0 output. In the follow-up release, editable text and movable image layers come directly out of inference, returning the model’s output as a stack of components rather than a flat frame. Soon after, the model will generate branded assets that follow a customer’s typography, colour palette, and logo without manual cleanup.<\/p>\n This is a fundamentally different proposition to “here is a nice image.” If the layer roadmap ships as described, Ideogram becomes a design tool, not just an image generator.<\/p>\n The weights are available on GitHub and Hugging Face under a commercial license. The nf4 quantisation fits on a single 24GB GPU. You accept the licence gate on the Hugging Face model page, create an access token, authenticate with the CLI, clone the GitHub repository, and install the package. Safety screening at inference is handled through Hive moderation with text and visual content moderation keys exported as environment variables.<\/p>\n The model is also integrated with the diffusers library. The remote prompt upsampling path gives the best results, while a local upsampling path uses the same Qwen3 VL 8B model as the text encoder for fully offline workflows, at some quality cost.<\/p>\n The open-weight release means you can fine-tune Ideogram 4.0 on your own brand and product data and deploy it inside your own environment. The weights, the fine-tunes built on top of them, the training data, and every inference call stay on your infrastructure. No data leaves your server. No API rate limits. No per-image pricing at scale.<\/p>\n For small studios and brands that have been priced out of training custom image models, this is the first open-weight option that produces output good enough for client work. The combination of text rendering, layout control, and colour conditioning means you can generate packaging mockups, social templates, ad creatives, and marketing collateral that would previously have required a closed API with per-call costs adding up fast.<\/p>\n The layer-based generation roadmap is the thing to watch. If editable text layers and branded asset generation ship as planned, this model becomes the foundation for an entirely different kind of design workflow: one where the AI does not just produce a final image but hands you the components to keep working with.<\/p>\n Want more like this? Join MOKU Club for free. Weekly resources, early access to new guides, and occasional templates you can actually use. Join below.<\/p>\n","protected":false},"excerpt":{"rendered":" Open-Weight Image Generation Just Caught Up to the Closed Models Ideogram 4.0 arrived on June 3 with a claim that sounded unlikely a month ago: an open-weight image model that…<\/p>\n","protected":false},"author":8,"featured_media":5187,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"footnotes":""},"categories":[71,64,60],"tags":[],"class_list":["post-5188","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai","category-design","category-news"],"yoast_head":"\nThe JSON Prompting Interface<\/h2>\n
2K Output, Any Aspect Ratio<\/h2>\n
Layer-Based Generation: Not Just a Flat Frame<\/h2>\n
Running It Locally<\/h2>\n
What This Unlocks<\/h2>\n