Black Forest Labs launches Flux 3 Image with multi-step editing. The new model changes one part of a picture at a time and leaves the rest alone, with layouts set by bounding boxes and output up to 4K. Black Forest Labs has released Flux 3 Image, the image side of its Flux 3 model family. It handles text-to-image, image-to-image, text rendering and photorealism, and the company's main claim is about editing: you can make several changes to a picture without anything else in it moving. Editing one part at a time. The point is consistency: changing a jacket color should not also bring a new face, a new background or different light. Black Forest Labs' product page describes making multiple edits at once without changing anything else, and gives recoloring, swapping and replacing single items as examples. Each change can be aimed at one area of the image while the rest is kept as it was. For anyone producing product shots, ad variations or a series of images that must stay consistent, this is the part to test first. If the claim holds up in your own tests, a change that leaves the rest of the frame alone means fewer rounds of regenerating and comparing. Laying out a scene with boxes. Instead of describing where things go in words and hoping, you can place elements with bounding boxes. The product page says positions are set on a grid from 0 to 1000 that works the same at any aspect ratio, and edits can be pointed at a specific box. You can also give the model up to ten reference images. Each one gets its own name in the prompt, such as ref_image_0 and ref_image_1, so you can say which reference supplies the product, which the person and which the style. Output goes up to 4K, and Black Forest Labs shows an example of 5456 by 3072 pixels straight from the model. Its examples also include legible text on signs, logos and printed material. Price and licensing. API access is 50 percent off through October 8. The company sells its models by the image through API credits, with no subscription and no free tier, so the free demo on its site is the way to try Flux 3 Image before paying. Companies can license commercial weights to run and fine-tune the model on their own hardware, which matters for teams that cannot send images to an outside API. An open-weight version is expected in the coming weeks. Ideogram is moving the same way. Shortly before this launch, Ideogram announced its own editing-focused model, version 4.5, which is also set to ship as open weights soon. If you edit images more than you generate them from scratch, both are worth comparing once they are out. Tools mentioned

Ideogram's P-Image drops per-image pricing to $0.003 with native 2K output. Ideogram launched P-Image, a family of image models co-developed with PrunaAI, according to the company's announcement on X. The company says pricing starts at $0.003 per image, the models render native 1K and 2K output, and generation lands in 3 to 8 seconds. P-Image targets teams generating images in volume, where speed and per-image cost weigh more than they do for one-off work. Four quality modes and native 2K, co-developed with PrunaAI. Ideogram ships P-Image as a model family with four quality modes. The modes let a team trade generation speed against output fidelity depending on the job. Native 1K and 2K output means the higher-resolution renders come straight from the model rather than an upscaling pass. The company built P-Image with PrunaAI, a firm focused on model compression and inference optimization. Ideogram credits that collaboration for the latency range it advertises, from 3 to 8 seconds per image. $0.003 per image undercuts the cost of iterating at scale. Ideogram says P-Image pricing starts at $0.003 per image. At that rate, a thousand generations runs about $3, which changes the math for workflows that produce many variations before a shot locks. Concept artists, previs teams, and marketing departments running large batches feel per-image cost more than anyone. Ideogram has expanded access across its lineup before. VP Land covered the company open-sourcing its Ideogram 4.0 weights for self-hosting and fine-tuning. Controls for production: JSON prompting, layout control, and text rendering. Beyond price and speed, Ideogram lists a set of controls aimed at repeatable output: * JSON prompting. Structured prompts let a team specify parameters programmatically, which helps when generation runs through a pipeline rather than a chat box. * Layout control. Direct placement of elements within the frame, useful for composition-driven work like posters and thumbnails. * Multiple aspect ratios. Support for different framing without cropping in post. * Photorealism, text rendering, and style versatility. Ideogram has built its reputation on legible in-image text, and it lists that alongside photorealistic and stylized output here. Available now across the API, Replicate, Leonardo, and Cloudflare. Ideogram says P-Image is available through its own API and a set of third-party platforms, including ComfyUI, Replicate, Leonardo, Cloudflare, and others. Distributing through the platforms where AI image workflows already run lowers the friction of testing the model against an existing setup. For studios weighing image models on cost per iteration, P-Image gives them a $0.003 floor at 2K resolution to measure against. The claims on quality and speed come from Ideogram; the test that matters is how the four modes hold up on a real production batch. TOOLS IN THIS STORY

How&How creates new identity for AI image platform Ideogram. The studio was tasked with distinguishing the platform from the sea of AI sameness by putting human taste, judgement and craft at its centre With so much AI slop out there, how do you set about rebranding a generative AI company that is committed to raising the standards of AI design, not lowering them? That was the question posed to branding agency How&How when Ideogram approached them to create a new identity for the platform. The resulting work tackles one of the most divisive conversations in contemporary design head-on - and it was apparently approached with a mixture of curiosity and caution. According to a recent Figma study referenced by How&How, 76% of professional graphic designers reported using AI tools in their workflow as of 2026 (up from just 35% in 2023). Yet 78% of UK designers said AI-generated work feels homogenised and undermines originality. As such, and given how high tensions are running right now around all things AI, it is perhaps unsurprising that the studio were initially reluctant to take on the work. How&How say the brief sparked extensive debate among the team, including a discussion on its internal 'moral compass' channel, where employees debate the ethical implications of potential work. "Working on a generative AI design tool is not a neutral brief for a branding studio," they admit. "But we also believe that understanding new technology means getting close to it, not observing it from a distance." Rather than presenting AI as a replacement for designers, the identity frames Ideogram as a tool intended to enhance creativity, while also underlining the importance of human judgement. At its centre is a stylised brain shape that flips the brand's previous logo, with a subtle 'I' embedded in the negative space in the middle as a reminder that ideas and decision-making still originate with people. Across the wider system, AI-generated imagery is combined with traditional creative tools and processes. Visuals appear in a constant state of motion, with particles swirling and synapses firing around the central cranial motif. Elsewhere, typography and abstracted visuals of animals, vegetation and everyday human-man objects are designed to emphasise precision and craft. The project arrives at a time when many AI companies are seeking to distinguish themselves from competitors offering broadly similar capabilities. To its credit, Ideogram seems to be pitching its tools to creatives who care as much about quality as having the ability to generate work with AI. Some may still take some convincing, but this refresh does a good job of communicating that distinction. And for How&How, the experience appears to have reinforced a conclusion many designers have already come to, and may find reassuring: yes, AI can help to accelerate the creative process, but it still takes human know-how, instinct and judgement to make something worth looking at.