
November 16, 2026 • 10 Min Read
November 18, 2026 • 12 Min Read

Melius carries one brand kit across every model it uses for AI image generation, Adobe Firefly trains custom models on your own assets, Midjourney builds a house look from style references and moodboards, Recraft pairs a custom style with your hex palette, Canva applies a Brand Kit to team designs, ChatGPT reuses saved reference photos, and Bria fine-tunes a model on your brand through its API.
The real difference between them is where your brand lives. In some tools it's a set of images you attach to each job. In others it's a saved style, a kit of exact colors and fonts, or a model that has your past work trained into it. That choice decides how much setup happens once and how much gets repeated on every image.
We compared the seven on four things: how the brand gets into the tool, whether colors and type come out right, whether the look holds across a whole campaign and everyone working on it, and the terms behind what you make.
Every tool on this list uses at least one of these approaches, and several combine two or more.
You attach pictures of the product, the person, or the look, and the model works from them for that generation. There's nothing to set up, but the references have to come along every time, which is where drift creeps in on a long campaign. Saved reference libraries close that gap. ChatGPT's named references and Melius Elements both store references once so they can be called up again.
A style reference passes along palette, lighting, and texture without copying what's in the picture. Saving it as a style, a code, or a moodboard means the next person on the team gets the same look without digging for the original image.
A brand kit stores the fixed facts of an identity: hex codes, typefaces, and logo files. It does the most for color and text, because the values are exact.
Training a model on a few dozen of your own images builds the style into the model, so prompts don't have to describe it. It's also the most setup of the four, and the trained model stays with your account.
Tool | How the brand gets in | Color and type control | Where it runs |
Melius | Brand kit read from your site or guidelines PDF, plus saved Elements | Up to six named hex colors and uploaded brand fonts | Web and desktop app, MCP, API, and CLI |
Adobe Firefly | Custom models trained on your images, plus style and composition references | Style and color follow the training images and references | Firefly web app, plus Photoshop and Adobe Express |
Midjourney | Style references, moodboards, and personalization profiles | Palette carried by the reference image | Web app and Discord |
Recraft | Custom styles from up to 10 reference images | Saved hex palettes alongside a style | Web app and API |
Canva | Brand Kit, Brand Templates, or a website URL | Brand colors, fonts, and logos | Web, desktop, and mobile apps |
ChatGPT | Reference images in chat, saved under a name | Described in the prompt | ChatGPT apps and the OpenAI API |
Bria | A model fine-tuned on your brand's images | Learned from the training set | API, no-code console, and the Bria Create app |
Melius is the AI-native operating system for creatives, where you direct AI agents to build your creative work. A brand kit holds a brand's voice, colors, typefaces, logos, and reference media, read from the brand's website, a guidelines PDF, or both. Once it's approved, the Melius agent applies it on every canvas, whichever of the many leading models available on Melius runs the image, such as Nano Banana Pro, GPT Image 2.5, or Flux 3 Image.
The Elements feature covers what repeats inside a campaign, such as a product, character, or brand ambassador. Upload label close-ups and dimensions with reference photos, and they will be reliably and accurately reproduced. A General Element can hold a campaign's mood or setting, and Melius picks the references each model has room for. To use one, mention it in a brief to Mel with #, up to three Elements per node.
What it does well. Teams keep one shared kit per brand and can generate with it across leading models. Elements keep key details consistent across different shots and angles. A new kit builds from a website or guidelines PDF in a few minutes, and the team's default kit applies to every canvas and every new chat, so an edit to a color, font, or logo carries into the next prompt. On Professional plans and above, you can reject a generated view of a product or redraw a wrong detail before the Element goes into a campaign.
The trade-off. The kit is only as good as its source, so review it before approving.
Best for. Agencies and in-house teams running several brands or campaigns across models.
Adobe Firefly is Adobe's generative AI app, with Adobe's own Firefly models alongside partner models from Google, OpenAI, and others. Custom models, in public beta since March 2026, let you train a private model on 10 to 30 of your own images to capture a style, a character, or a photographic look.
What it does well. Style reference matches the look of an uploaded image, and composition reference keeps its outline and depth, so a campaign's style and its framing can both stay fixed. Adobe trains its own Firefly models on licensed and public-domain content. Enterprise accounts can train custom models on brand assets too and use them in Photoshop and Adobe Express.
The trade-off. Custom models are still in beta, and Adobe says they're tuned for ideation. Adobe's help pages also say its composition, style, and color controls work only with Adobe's own models, not the partner ones.
Best for. Creative Cloud teams that want their own trained style inside the apps they already use.
Midjourney makes images with its own models, through its website and a Discord bot. Its brand tools are about the look of an image. A style reference carries the colors, medium, texture, and lighting of an image, a moodboard turns a set of chosen images into one style, and a personalization profile learns your taste from the images you like or select.
What it does well. A style reference image or a saved moodboard can go into any later prompt, so a look carries from one campaign image to the next. In V8.2 the Edit Model takes up to four reference images, replacing Omni Reference and Character Reference.
The trade-off. Midjourney doesn't have a brand kit for hex codes or fonts, so exact colors and type depend on the prompt and a careful review. A style reference also won't carry over the objects and people in the picture, so a product or a logo needs its own reference.
Best for. Art directors defining a recognizable visual style for a brand.
Recraft makes raster images and vector graphics you can keep editing. A custom style is built from up to 10 reference images (in V4 styles) and saved with a palette setting, so the look and the colors travel together into every project.
What it does well. Palettes can be typed in from a brand guide's hex codes, sampled from an image with the color picker, or pulled from an uploaded picture. Vector output keeps logos, icons, and illustrations editable after they're generated.
The trade-off. Recraft's own guide says a reference doesn't guarantee exact colors, so the saved palette does the real work on color. On the free plan, Recraft owns what you make, shows it publicly, and doesn't license it for commercial use.
Best for. Identity and illustration work where assets have to stay editable as vectors.
Canva is a design platform with AI generation built in. Its Brand Kit stores a team's colors, fonts, and logos, and Brand Templates added to the kit show up inside Canva AI when it generates a presentation or a social design. A team without a kit can have Canva AI pull colors, fonts, logos, and imagery from a website URL instead.
What it does well. The kit sits at the team level, so everyone designs from the same palette and type, and brand designers and admins choose which templates the AI draws on.
The trade-off. Canva AI doesn't support Brand Controls and Permissions yet, so protecting a template means using editing locks. It generates fixed-size designs such as presentations and social posts, not websites or video.
Best for. Teams producing on-brand social posts and presentations every week.
ChatGPT generates and edits images in conversation. Images 2.6, released on November 4, 2026, lets you save a reference photo under a name and call it into a later chat by typing @ and that name.
What it does well. Edits work by plain instruction, and OpenAI says Images 2.6 keeps details more consistent from one edit to the next. Through the API, a single edit can combine several reference images into a new image.
The trade-off. There's no brand kit, so color and type rules go into the prompt every time. OpenAI's own image guide lists keeping recurring characters and brand elements the same from one generation to the next as a known limitation.
Best for. Individuals making quick on-brand edits without opening another tool.
Bria builds image models on licensed data and sells them to businesses through an API, a developer platform, and its Bria Create app. Tailored Generation fine-tunes Bria's model on a company's own assets, such as a mascot, a product look, or a brand style.
What it does well. Training can be fully automatic or hands-on for teams that want to set the parameters, through Bria's console, its API, or an embedded iframe. Bria trains with data isolation, offers enterprise indemnity, and can deploy a finished model in a public or private cloud or on premises.
The trade-off. Tailored Generation is aimed at enterprise teams, and getting value from it takes a project's worth of training, testing, and deployment.
Best for. Platforms and enterprises building brand-tuned image generation into their own products.
Start with the brand kit. Give Melius your website, add the guidelines PDF if there is one, check the colors, type, and logos it found, and approve the kit. Then save the campaign's recurring pieces as Elements, such as a product Element for the hero product and a General Element for the season's look.
From there, brief Mel the way you'd brief a designer and mention each Element with #. A spring launch might read: "Six images for the #Citrus tonic launch, two square feed posts, two 4:5 ads, and two 9:16 stories, in the #Spring campaign look. Keep the label wording exact and set any headline in our brand type." See how Elements store products and visual styles.
Mel lays out a node for each image with the kit applied. When one comes back off, swap the model on that node or rerun it with a tighter reference, and the other five stay as they are. Identity, packaging, and campaign systems run on the same canvas, covered on Melius for branding.
Running more than one brand through AI image work? Build a brand kit on Melius and let Mel apply it to the next campaign.

November 16, 2026 • 10 Min Read

November 15, 2026 • 9 Min Read

November 14, 2026 • 8 Min Read