What actually changed in October 2026
If you build recurring characters with AI image or video tools, October 2026 was a genuinely big month. Several of the major platforms creators rely on for consistent casts updated how they handle character references, and the shift was large enough that a widely cited character-consistency benchmark explicitly flagged its own April 2026 test results as outdated.
The changes landed across the board. ChatGPT now accepts uploaded reference images. Midjourney V7 replaced its old --cref parameter with a new system called Omni Reference. Ideogram added a single-image character feature. Google's Gemini image models, Nano Banana 2 and Nano Banana Pro, now take several character reference images at once instead of just one.
The common thread is that every one of these tools moved toward the same basic idea: give the model one or more clean reference images of your character, and let it hold that character's likeness across new poses, scenes and styles. That's a meaningful departure from the older approach of re-describing a character in text every time you wanted a new shot.
Midjourney, Ideogram, and Nano Banana, tool by tool
Midjourney's change is the most structural. If you were using --cref, that parameter belongs to V6. From V7 onward, Midjourney's reference tool is Omni Reference, invoked with --oref and controlled with a weight parameter, --ow. If you've still got old --cref prompts saved, they won't carry over cleanly, so it's worth rebuilding your reference workflow around Omni Reference now rather than patching old prompts later.
Ideogram's addition is called Ideogram Character. It keeps one character recognizable across new poses, scenes and styles from a single reference image, and notably requires no model training. That makes it one of the lower-friction options on this list: you upload one good image and go, rather than assembling a multi-angle set.
Google's Nano Banana 2 and Nano Banana Pro take a different approach by accepting several character reference images at once rather than just one. That multi-image input is closer to what Krea 3, another October 2026 launch, offers: native 2K edits in under 8 seconds, at $0.06 an image, with up to 6 reference images to keep products and characters consistent across a campaign.
Video models are catching up too
This isn't just an image-generation trend. Video models are adopting the same reference-based logic, and in some cases going further. Kling 4.0's Omni Reference takes up to 15 references at once, counting images, videos, voice profiles and elements together as a single pool. Kling explicitly names three-view character sheets, grid images, storyboards and wireframes among the image types it accepts, which matters a lot if you're trying to keep the same character consistent across multiple shots in a sequence.
Other video tools are converging on reference images as the input method too, even if the exact mechanics differ. Google Flow, Kling and Seedance all take reference images when they generate video. Runway works a little differently: its References feature produces consistent stills, which creators then animate separately rather than feeding straight into video generation.
The practical upshot is that the vocabulary and workflow of character references, once scattered across incompatible per-tool parameters, is starting to look similar whether you're generating a still image or a video clip.
Why a character sheet beats a text description
The guidance that's emerged alongside these updates is straightforward: stop describing your character from scratch in every single prompt. Text descriptions drift. A character described as having short reddish hair and a green jacket will come out slightly different every time you type that description into a new prompt, because the model has to reinterpret the words each time.
A character sheet, by contrast, gives every one of these tools something stable to anchor to. Build one clean reference image, or two to three reference angles if a tool supports multiple images, and reuse that same source material across whichever platform you're working in that day, whether that's Midjourney's Omni Reference, Ideogram Character, Nano Banana, or Kling's Omni Reference.
This matters more now than it used to, precisely because all the major platforms converged on this same reference-image approach in the same month. Cross-tool character consistency is easier to achieve today than it has been at any previous point. But that only holds if the reference images you feed in are clean and consistent to begin with. A blurry, badly lit, or inconsistent source image will produce inconsistent results no matter how advanced the reference tool is.
- Build one strong reference image, or 2-3 consistent angles if the tool accepts multiple
- Keep lighting, pose clarity, and framing consistent across any multi-image reference set
- Reuse the same source images across tools rather than regenerating character descriptions per platform
- Update old Midjourney prompts from --cref to --oref and --ow if you're on V7
What this means for your workflow
The broader context here is a growing interest in persistent AI casts, characters creators can reuse across episodes, scenes and art styles without rebuilding them from zero each time. Dedicated cast products now let creators describe a character in a line of text, or add a photo in the app, and get a cast member who stays recognizably themselves across every scene. Episodo is one tool built around that exact workflow, for creators who want a cast they can reuse episode to episode without re-engineering character consistency by hand.
Whether you use a dedicated cast tool or you're stitching together Midjourney, Ideogram, Nano Banana and Kling yourself, the standardization happening this month is worth taking seriously. The brief doesn't specify exactly how these different reference systems compare in output quality against each other, so it's worth testing your own character sheet across two or three tools before committing to one as your primary pipeline.