Does Gemini Omni 1.2 Flash need a turnaround sheet?

A turnaround helps. The launch post says the model follows character sheets into consistent video, because it can hold a character's face, clothes and product details across camera angles. A front view alone leaves the side and the back for the model to invent.

Should I upload the whole sheet, or separate views?

Separate files. One sheet image contains every angle, so the prompt cannot point at image 2 as the side view. Front, side, back, full body, prop and scene are six images. The limit is 10, so four slots remain.

Why is a first-and-last-frame clip capped under 15 seconds?

Generation and editing run from 3 to 15 seconds. First-and-last-frame mode is 3 to 10 seconds on the Cloud model card. For a 15-second clip, generate from references without locking both end frames.

Gemini Omni 1.2 Flash reference to video with a character pack

Gemini Omni 1.2 Flash (model ID gemini-omni-1.2-flash) generates a clip from reference images you attach separately. Google's 5 November 2026 launch says it follows character, product and style images more closely, and specifically calls out turning character sheets into consistent video. This page maps a turnaround pack onto that image limit and shows a plain-language prompt for each upload.

We have not made an example with Gemini Omni 1.2 Flash yet. This page maps the pack to Google's published limits; the examples on this site were made with Kling 3.0.

Gemini Omni 1.2 Flash (official limits)

Checked 2026-11-06

Generally available on 5 November 2026 on the Gemini API and Google Cloud. Model ID gemini-omni-1.2-flash. The Interactions API is supported. The GenerateContent API is not supported for this model.
Maximum 10 images per prompt. Inline or console files up to 20 MB each; files from Google Cloud Storage up to 30 MB. MIME types: image/png, image/jpeg, image/webp, image/heic, image/heif. Maximum 3 videos per prompt.
Generate or edit from 3 to 15 seconds, in 1-second steps (Omni 1.1 stopped at 10 seconds). First-and-last-frame mode is 3 to 10 seconds. Aspect ratios 16:9 and 9:16. Resolutions 360p, 720p, 1080p and 4k. The reference-to-video docs default to 720p when resolution is omitted.
About 30% less than Omni 1.1: $0.07 per second for 720p video, with similar savings at other resolutions. Draft at 360p or 720p, then upscale a clip you keep to 1080p or 4K.
Guide a clip with character, product or style images, and with short motion clips of up to 3 seconds. The launch post also mentions audio cues. The Cloud blog says audio references are behind an allowlist, and the model card lists audio as an input modality as not supported. Generated speech, music and sound effects are listed as supported.
Reference-to-video uses the task reference_to_video. Attach each image as its own input. The documented sample prompt is: "Use the image(s) as references to generate the video."

Sources: Google: Introducing Gemini Omni 1.2 Flash · Google Cloud model card: Gemini Omni 1.2 Flash · Google Cloud: Generate videos from references · Google Cloud blog: Gemini Omni 1.2 Flash generally available · Google DeepMind model card: Gemini Omni 1.2 Flash

Upload order

Upload one view per file, in this order, and name that order in the prompt. Six images leave four slots free under the 10-image limit.

image 102_turnaround_views/view_front.pngFace and front of the outfit
image 202_turnaround_views/view_side.pngProfile and hair from the side
image 302_turnaround_views/view_back.pngHair and outfit from behind
image 401_character_reference/full_body.pngHeight and proportions
image 504_props/prop_role.pngThe product or tool
image 605_scenes/scene_workspace.pngThe location

The hard limit on the Cloud model card is 10 images per prompt. Google's examples often use a small set of multi-view character or product shots for identity. That is a choice of how many to send, not a lower cap. Use a spare slot for a second character, an outfit or a detail crop only when the shot needs it.

Why three separate views beat one good angle

A single image, even a well-chosen three-quarter view, only shows part of the character. When the character turns, the model has to invent the profile and the back, and that is where the face, hair and outfit start to drift. Upload the front, side and back as separate references and give each one a single role in the prompt, as the official guide recommends. If you also have a three-quarter view, add it as an extra reference rather than using it on its own.

Keep one face per image: crop expression grids out of the sheet, and paste the same locked appearance description into every shot's prompt.

Prompt template: one character

Prompt template

Use the uploaded images as references to generate the video. Image 1 is the front view of [Name], image 2 is the side view, and image 3 is the back view. Keep the same face, hairstyle and outfit in every frame. Image 4 sets body proportions. Image 5 is the [product]; keep its shape and colors. Image 6 is the location; keep its layout and lighting. [Name] is in [location]. [Action]. [Camera], one continuous shot, 16:9, 8 seconds. No extra people, no subtitles, no logos.

Write the role of each image in plain language, tied to upload order. The docs do not require an image-index tag.

Edit the same clip instead of starting over

A follow-up turn can change the clip you already have: swap the product or the background, change the lighting or the weather, or switch the aspect ratio. Google says this works on a full 15-second clip. For a character ad, keep the face, outfit and action, and change only the line you name.

Prompt template

Keep this same clip and the same character. Do not regenerate the performance. Swap the background for [new place] and change the lighting to [lighting]. Keep [Name]'s face, hairstyle, outfit, action and camera framing. Leave the duration and 16:9 aspect ratio unchanged.

On the Interactions API, set store to true if you want to reference that interaction for a later edit. Stored interactions are kept for up to 7 days. The Cloud blog also lists changing the camera angle as an edit, rather than generating the shot again.

Omni 1.2 Flash, Kling 4.0, or Seedance 2.5

All three can take separate front, side and back views from the same pack. Omni 1.2 Flash allows 10 images, clips of 3–15 seconds, aspect ratios 16:9 and 9:16, and $0.07 per second at 720p, and a later turn can edit that same clip. Kling 4.0 Omni Reference also allows 10 images, inside a 15-item reference total, with clips of 3–30 seconds and more aspect ratios; the Kling guide on this site lists $0.157 per second at 720p without an input video. Seedance 2.5 allows up to 30 images and clips of 4–30 seconds; its guide lists a 720p example of $0.231 per second without an input video. Choose by length, frame shape and price, then reuse the same pack files.

Kling 4.0 Seedance 2.5

Questions

Does Gemini Omni 1.2 Flash need a turnaround sheet?: A turnaround helps. The launch post says the model follows character sheets into consistent video, because it can hold a character's face, clothes and product details across camera angles. A front view alone leaves the side and the back for the model to invent.
Should I upload the whole sheet, or separate views?: Separate files. One sheet image contains every angle, so the prompt cannot point at image 2 as the side view. Front, side, back, full body, prop and scene are six images. The limit is 10, so four slots remain.
Why is a first-and-last-frame clip capped under 15 seconds?: Generation and editing run from 3 to 15 seconds. First-and-last-frame mode is 3 to 10 seconds on the Cloud model card. For a 15-second clip, generate from references without locking both end frames.

Browse character turnaround sheets Character reference sheet prompts

Start with a free character

Every character's full 19-image pack is free: eight starter characters need no account, the rest a free sign-up. All nine example ads use pack images only.

Browse free characters