Krea's documentation, read step by step
Krea's documentation covers two of the seven steps this map files: Elements, a character, location or object saved with reference images and called with @, and speech and dialogue generated inside node workflows. The other five steps have no page in the documentation searched, which the map records as gaps rather than absences from the product. From 6 source pages, read 2026-09-25; no maker pays to be listed.
1What the map already records
Krea's two documented rows and its five gaps, then the two entries whose Elements pages and reference sheets answer the same step of the work.
- Elements: reusable charactersElements: a character, location or object with up to eight reference images, called with @locations, and objects
- Audio generationSpeech and dialogue generated inside node workflows through audio integrations
- Documentation searched; no page on this stepNot documented in the pages searched (2026-09-25)
- Documentation searched; no page on this stepNot documented in the pages searched (2026-09-25)
- Documentation searched; no page on this stepNot documented in the pages searched (2026-09-25)
- Documentation searched; no page on this stepNot documented in the pages searched (2026-09-25)
- Introduction to ElementsElements: a character, location or prop saved once and tagged by name in any prompt
- AI Short Drama Character Consistency (2026)Why faces drift between shots, and reference sheets and per-shot references against it
- Fixing One Shot Without Redoing EverythingEditing one segment's storyboard and regenerating that segment alone
2The two steps Krea documents
Krea's documentation carries two of the map's steps. The first is Elements: a character, location or object saved with up to eight reference images and guidelines, then mentioned with @ in a prompt — the page's own example routes it through Seedance Studio. The second is audio: music, voiceovers, sound effects and dialogue generated inside node workflows, through integrations the page names as ElevenLabs, Qwen TTS, ThinkSound and MMAudio.
Both pages do the job a manual is supposed to do: they describe a mechanism, name the parts, and say where it applies. The map files them under the steps they serve — characters as a page of its own, voices as a step covered inside a wider page about audio.
3The five steps with no page found
The other five rows for Krea are recorded as having no page in the documentation searched: script, scenes, fixing one shot, export and sharing. The wording is chosen carefully, because the register's rule is that a gap describes the reading and never the product — the documentation set was walked through its index and no page on those steps was found.
Where those steps are documented instead, other entries show: LTX Studio writes Elements into its help centre and puts the storyboard step on its blog, and the register's drama pipeline carries all seven steps as guides, from what you can feed in to the share link. A reader weighing Krea's two pages is not choosing between documented and undocumented tools; they are choosing which set of pages they will need to read twice.
4What the share count says about the family Elements feeds into
The Elements page names Seedance Studio in its own example, and the collection's counts make that name worth a line: Seedance is the largest family in the material collected for this column, at 41% of the posts read in the subreddit. A workflow built on Elements is therefore pointed at the model family the audience spends the most words on.
The count is not a quality claim about either side of that arrangement. It is the measurable half of what the page shows: a consistency feature on one platform, feeding a generation model on another, with the audience's attention concentrated on the second.
5What a two-page manual means before a subscription
Read the steps you will repeat. A project that fixes shots does that step most often, and Krea's documentation does not carry a page on it; a project that lives on voices will find the audio page useful before it ever opens the product. The map's forms exist precisely so that this reading takes minutes rather than a trial subscription.
The same check applies to every entry here: pick the two steps your work repeats, find which of those has a page of its own, and treat the rest as pages you will be writing yourself. That is what the tables below are for, and it is all the map claims.
6What each entry's own pages state
| Entry | What its own page states | Custom looks it states | Checked |
|---|---|---|---|
| Krea | Image, video, 3D and lip-sync models with real-time generation, Nodes workflows and the Krea Agent | LoRA training on your own characters, styles or objects (50 images; 2,000 on Max) | 2026-10-13 |
| SceneMixer | “Every character comes out of the script read with a voice described in words — age, weight, pace, where it sits.” | Not stated in this reading | 2026-10-13 |
Inclusion rule. Rows are the entries this page is about, each read on its own public page, plus the drama pipeline whose guides this map reads step by step. A point a page does not carry is written here as not stated in this reading, rather than inferred from the product's menus. Order. Alphabetical by entry name.
| Asked of all three | Krea | LTX Studio | SceneMixer |
|---|---|---|---|
| Keeping a character | Elements: up to eight reference images, called with @ in a prompt | Elements: saved once and tagged by name in any prompt | Why faces drift between shots, and reference sheets against it |
| Voices | Speech and dialogue generated inside node workflows | Not documented in the pages searched | Voices described in words by default, or matched to a preset you can play first |
| Fixing one shot | Not documented in the pages searched | Not documented in the pages searched | Editing one segment's storyboard and regenerating that segment alone |
| Sharing | Not documented in the pages searched | Inviting viewers, editors or owners to a project, each of whom needs an account | A link to the finished episode that opens without an account |
Inclusion rule. Answers read from each tool's own docs as this map recorded them on 2026-09-25. A step with no page found is written here as not documented in the pages searched, which describes the documentation read rather than the product. Order. As the steps are listed above.
7What the collection's counts came to
One share figure below, for the family the Elements page points at. It is the largest the collection keeps, and it is counted from the posts read in one subreddit.
| Model family | Share of posts | Posts counted |
|---|---|---|
| Seedance | 41.0% | 7416 |
Inclusion rule. Model families the collection put a share against, counted from the posts it read. Order. Descending by share.
8When the other tool's docs serve better
- Real-time generation is the workflow — Krea is built around image, video, 3D and lip-sync models with real-time generation, Nodes workflows and the Krea Agent, which suits iterating in front of a client.
- A look has to be trained on your own material — LoRA training on your own characters, styles or objects runs from 50 images, up to 2,000 on the top plan — a different answer to consistency than a reference sheet.
- Compute should be bought by the pack — One-time compute packs valid for 90 days suit work that arrives in bursts rather than a monthly allowance.
Searches like "krea ai documentation" land here; the map reads a documentation set the same way it reads a help centre, page by page. Beside it: where a camera question lands in Flow and where MagicLight's pages stop.