Image (Keyframe) Generation

Cutflow's image generation starts from the script. The AI analyzes the script to write prompts and references Character Sheets and location images to generate high-quality images featuring consistent characters and backgrounds.

Overview

In Cutflow, image generation is not simply about entering text to create images. Through a script-first workflow, the AI understands the context of the script and references registered characters' Character Sheets to generate images with consistent characters.

Generated images are called Keyframes, and they serve as the reference point for subsequent video generation. Selecting a representative image from 4 candidates and generating video based on it is Cutflow's core workflow.


How Image Generation Works

Script-First Workflow

Cutflow's image generation starts from the Script.

Write Script  ->  AI Auto-Generate Prompt  ->  Review/Edit Prompt  ->  Generate 4 Images

Step 1: Write the Script

Describe the scene in the cut's script area. Include @charactername and @locationname in the script to specify appearing characters and locations.

@Seohee is sitting by the warm cafe window, cradling a coffee cup in both hands.
Afternoon sunlight filters gently through the glass, and Seohee gazes outside with a smile.

Step 2: AI Prompt Auto-Generation

Click the [Generate Prompt] button, and the AI (Gemini 2.5 Flash) analyzes the script to write a prompt optimized for image generation.

What the AI does:

  • Extracts visual elements (location, characters, actions, mood) from the script.
  • Recognizes @character and @location mentions and automatically includes their descriptions in the prompt.
  • Structures the prompt in a format suitable for image generation models.

Cost: AI prompt generation is free (no credit cost).

Step 3: Review and Edit the Prompt

Review the generated prompt and edit if necessary.

  • You can freely edit the prompt.
  • You can add or remove @character and @location mentions.
  • You can also write a prompt from scratch (without AI auto-generation).

Step 4: Generate Images

Click the [Generate Images] button, and 4 candidate images are generated using the selected model and reference images.

What Cutflow does during generation:

  1. Detects @character and @location mentions in the prompt.
  2. Automatically attaches each character's selected variation's Primary Character Sheet and each location's selected variation's Primary image as references.
  3. Attaches any additional reference images after the character/location references.
  4. Applies the aspect ratio matching the project settings.
  5. Generates images using the selected AI model.

@Mentions and Reference Images

Automatic Reference Attachment

When a prompt contains @charactername or @locationname, Cutflow automatically attaches the corresponding asset's image as a reference image. This is the core mechanism through which Cutflow maintains both character and background consistency.

Sequence of operations:

  1. Detects mentions like @Seohee, @Cafe in the prompt.
  2. Collects each character's selected variation's Primary Character Sheet and each location's selected variation's Primary image.
  3. Places them in the reference image list in mention order (characters first).
  4. Sends these reference images along with the prompt to the AI image generation model.

Important: Character reference images are always positioned at the very front of the reference list. This helps the AI model recognize characters more accurately.

Reference Images Panel

In the Cut Editor's Keyframes tab, the Reference Images Panel is displayed below the prompt input area. You can use this panel to review and manage the reference images that will be used for the current generation.

Each character/location has a variation dropdown, letting you select a different variation (outfit, time of day, etc.) to use for that specific cut only. For more details on the variation system, see the Variation System document.

Tip: The Shot Plans & Auto-Flight feature lets you automatically batch-generate keyframes for all cuts at once.

Panel layout:

[Character References]           [Additional References]
+--------+  +--------+          +--------+  +--------+
| @Seohee|  | @Jake  |          | Cafe   |  | + Add  |
| [Var ▼]|  | [Var ▼]|          | BG     |  |        |
+--------+  +--------+          +--------+  +--------+
  • Character references (automatic): Sheets for characters detected from @mentions are displayed automatically. You can switch variations via dropdown.
  • Additional references (manual): You can manually add/remove reference images for backgrounds, styles, etc.

Note: For more details on reference image management, see the Character Management > Reference Images document.

Variation Override

If you want to use a different variation instead of the character's default for a specific cut, you can switch the variation via dropdown in the Reference Images Panel.

  1. Click the [Variation ▼] dropdown on the character in the Reference Images Panel.
  2. All variations for that character are displayed.
  3. Select the desired variation.

Note: This change applies only to the current cut's generation. The character's default variation is not changed.


Keyframe Selection

Comparing 4 Candidate Images

When image generation is complete, 4 candidate images are displayed in a 2x2 grid on the Keyframe Canvas in the center panel.

Each image provides the following features:

FeatureDescription
Lightbox ZoomClick an image to view it at full size in the lightbox
Select as RepresentativeUse the [Select as Representative] button below the image to designate it as the representative keyframe
DeleteYou can individually delete unwanted images

Selecting the Representative Keyframe

Select the most suitable image from the 4 candidates as the representative keyframe.

Role of the representative keyframe:

  1. Displayed as the thumbnail in the left cut list.
  2. Used for the overall storyboard flow check.
  3. Used as the reference point (first frame) for subsequent video generation.

How to select:

  1. Review all candidate images (lightbox comparison recommended).
  2. Click the [Select as Representative] button on the preferred image.
  3. The selected image gets an orange border, and the cut status changes to Keyframe Generated.

Tip: You can change the representative keyframe at any time. Selecting a different candidate image deselects the previous one.

Lightbox Comparison

Clicking an image opens a lightbox where you can view the image at full size.

  • Navigate between candidate images using left/right arrows.
  • Zoom in to inspect character faces, background details, etc. in detail.
  • You can also select the representative from within the lightbox.

Regeneration

If none of the candidate images are satisfactory, click the [Regenerate] button to generate a new set of 4 candidate images.

  • You can edit the prompt before regeneration.
  • Previously generated candidate images are preserved (unless deleted).
  • Regeneration consumes additional credits.

Custom Keyframe Upload

Instead of AI-generated images, you can upload images created with external tools or made by hand as keyframes.

  1. Click the [Upload] button on the Keyframe Canvas.
  2. Select an image file.
  3. The uploaded image is registered as a keyframe and can be selected as representative.

Cost: Custom keyframe upload is free.


Advanced Settings

Model Selection

Cutflow supports various AI image generation models. By default, Cutflow automatically selects the optimal model for the project settings, but you can also change it manually.

Image Generation Model List

ModelCredits/ImageTierCharacter ConsistencyFeatures
Flux 2 Flash1FastweakLowest cost, for quick previews
Flux 2 Turbo2FastmediumFastest generation
Gemini 2.5 Flash2FastmediumGoogle model
Flux 2 Dev3StandardmediumLoRA custom support
Gemini 3 Pro4StandardmediumGoogle model
Flux 2 Pro (default)5StandardmediumDefault production model
InstantCharacter5StandardstrongZero-shot character consistency
IP-Adapter Face ID5StandardstrongSpecialized for face consistency
Flux Kontext Pro6PremiumstrongSpecialized for character editing
Recraft V36PremiumnoneVector art, text rendering
Flux 2 Flex8PremiumstrongMulti-reference, IP-Adapter

Default model: Flux 2 Pro (5 credits/image) is used unless you select otherwise.

Tier Selection Guide

TierCredit RangeRecommended Use
Fast1-2 cr/imagePrompt testing, quick composition checks, idea sketches
Standard3-5 cr/imageGeneral production work, final keyframe generation
Premium6-8 cr/imageKey scenes requiring highest quality, complex character interactions

Character Consistency Level Guide

LevelRecommended ModelsDescription
strongInstantCharacter, IP-Adapter Face ID, Flux 2 FlexScenes where character appearance is critical
mediumFlux 2 Pro, Gemini 3 ProGeneral character scenes
weak / noneFlux 2 Flash, Recraft V3Background-focused scenes, images with text

Generation Count Setting

By default, 4 candidate images are generated, but you can adjust the count between 1 and 4.

CountCredits (Flux 2 Pro basis)Recommended Use
1 image5 crWhen you need just 1 image quickly with a finalized prompt
2 images10 crWhen comparison is needed but you want to save credits
4 images (default)20 crSecure enough candidates for optimal image selection

Tip: Credits are charged per image. Generating 4 images costs the model's credits x 4. Example: Flux 2 Pro 4 images = 5 x 4 = 20 credits

Character Sheet Override

To use a different sheet instead of the character's default (Primary) sheet for a specific cut, you can specify the sheet via the character_sheet_overrides option in the image generation request.

In the UI, you switch sheets through the variation dropdown in the Reference Images Panel.

Additional Reference Images

In addition to Character Sheets and location images, you can attach additional reference images for backgrounds, styles, etc. Additional reference images are placed after the character/location references.

API transmission order:

[Character Sheet 1, Character Sheet 2, ..., Location Image 1, ..., Additional Reference 1, Additional Reference 2, ...]

Limit: You can attach up to 5 additional reference images.

Quality Preset

You can select the quality level of the generation model:

PresetDescriptionAvailable Models
fastFast generation, lower detailFast tier models
standardBalanced quality and speedStandard tier models
premiumHighest quality, slower generationPremium tier models

Step-by-Step Guide: Credit-Efficient Image Generation

Here is how to achieve desired results while using credits efficiently.

1. Test Composition with a Fast Model

  1. Set the model to Flux 2 Flash (1 cr/image).
  2. Set the generation count to 2 images.
  3. Test prompts to check composition and mood.
  4. Total cost: 2 credits (1 cr x 2 images)

2. Adjust the Prompt

  1. Review the images generated with the Fast model and refine the prompt.
  2. Adjust composition, character positioning, background elements, etc.
  3. Repeat until you are satisfied with the prompt.

3. Final Generation with a Standard Model

  1. Once the prompt is finalized, change the model to Flux 2 Pro (5 cr/image).
  2. Set the generation count to 4 images.
  3. Generate high-quality candidate images.
  4. Total cost: 20 credits (5 cr x 4 images)

Total Cost: 22 credits

With this method, you spend 2 credits on prompt testing and 20 credits on final generation for a total of 22 credits to secure a satisfactory keyframe.

Comparison: Generating 4 images with Flux 2 Pro 3 times from the start would cost 60 credits.


Credit Cost Reference

Image Generation Credits

ItemCredit Charge TimingReason
Image generationPre-charged before generationPrevents concurrent request overflow
Generation failureAutomatic refundCredits are returned for failed generations

Credit Calculation Formula

Image Generation Credits = Credits per image (by model) x Number of images

Examples:

  • Flux 2 Flash 4 images: 1 x 4 = 4 credits
  • Flux 2 Pro 4 images: 5 x 4 = 20 credits
  • Flux 2 Flex 2 images: 8 x 2 = 16 credits

When Credits Are Insufficient

When credits are insufficient, image generation is blocked and a credit recharge notification is displayed.

  • The credit badge in the top NavBar turns warning color (amber) when the balance is below 50.
  • Clicking the generate button displays the credit purchase modal.

Tips & Best Practices

  1. Test with a Fast model first: Use an inexpensive Fast model for checking composition and mood, not for final output.
  2. Write detailed scripts: The quality of AI prompt auto-generation is proportional to the specificity of your script.
  3. Ensure Character Sheet quality: The quality of the sheet determines the character consistency of all subsequent images.
  4. Compare in the lightbox: Do not judge the 4 candidates by thumbnails alone -- zoom in via the lightbox to compare details.
  5. Use strong models when character consistency matters: InstantCharacter and IP-Adapter Face ID are specialized for character appearance consistency.
  6. Use additional references for background consistency: For scenes repeating the same location, attach the previous cut's keyframe as an additional reference image.
  7. Edit the prompt manually: Rather than using the AI-generated prompt as-is, add desired elements or remove unnecessary parts to fine-tune it.

Frequently Asked Questions

Q. How long does image generation take?

It varies by model, but generally 4 images are generated within 10-30 seconds. Fast tier models are faster, and Premium tier models may take longer.

Q. Can I enter a prompt directly without a script?

Yes. You can skip AI prompt auto-generation and write your desired prompt directly in the prompt input field to generate images.

Q. Can I delete previously generated images?

Yes. You can delete individual images on the Keyframe Canvas. However, deleting an image selected as the representative keyframe will unset that cut's representative keyframe.

Q. What happens to previously generated videos when I change the representative keyframe?

Previously generated video Takes are preserved. However, when generating new videos, the changed representative keyframe will be used as the reference point.

Q. Can I use a Negative Prompt?

Yes. You can enter a Negative Prompt in the advanced settings to instruct the AI to exclude certain elements from the image. (e.g., "blurry, low quality, distorted face")

Q. Is the aspect ratio applied automatically?

The aspect ratio set during project creation (9:16, 16:9, 1:1, 4:5) is automatically applied to image generation. You cannot change the ratio for individual cuts.

Q. Can I generate images multiple times for a single cut?

Yes. You can use the [Regenerate] button to generate new sets of candidate images, and previous generation results are preserved. Select the best image from multiple sets of candidates as the representative.

Q. Are there models that do not support reference images?

Some models (e.g., Recraft V3) do not support reference images. When you use these models, images are generated from the prompt text alone. Use models that support reference images for scenes where character consistency is important.

Q. How many images can I generate with the Starter package ($10, 200 credits)?

With the default model (Flux 2 Pro, 5 cr/image), you can generate approximately 40 images (200 / 5 = 40). Using a Fast model (Flux 2 Flash, 1 cr/image), you can generate up to 200 images.


Related Documents