ModelBrief · Kling 4.0 Flash · Topic Page
Last updated Nov 20, 2026

Kling 4.0 Flash Guide: API, MCP and Omni Workflow

ModelBrief · Published 2026-11-18 · Fact-check 2026-11-16 · Reference: Kuaishou Kling official API docs and usage guide, OrcaRouter

Since its Nov 4, 2026 general availability, Kling 4.0 Flash has four routes in: the website and app, the official API, MCP/CLI, and the newly opened AI plugins. Flash's correct role is the draft tier — batching samples and validating ideas fast, with finals handed to the full Kling 4.0. This article walks through each route, then the Omni reference workflow, then the draft-vs-finals trade-off. Two unverified items first: the Adobe plugins rest only on Kling's X post via secondary sources, not verified against an Adobe listing; and Flash's native audio is disputed across sources, so do not assume sound in your workflow.

Route 1: website and app (fastest start)

Flash is open to all plans, billed in credits on the website. Text-to-video, image-to-video and Omni multi-reference all work in the browser; per-request caps are 20s and 720p. Good for a first try: no code, just run a few clips to see whether the style and motion fit. Note: no Flash-specific free allowance has been verified, so do not budget on a free tier; prices per the official pricing page.

Route 2: the official API (three paths)

Three official API paths, published with the Nov 4 GA: text-to-video, image-to-video, Omni multi-reference. Docs: 4.0 Flash text-to-video and the Kling 4.0 usage guide (§5.1). Billing: no input video, 0.6 Units/s (≈ $0.094/s), billed on output duration; with input video, 0.4 Units/s, billed on 2× output duration when input ≤ output — effectively ≈ $0.126/s. Confirm which row you are on before calling the API.

Prompt budget: Flash takes an English prompt of up to 26,000 characters (per OrcaRouter, from the official docs); the full model takes text prompts of up to 8,000 tokens. The two figures use different units (characters vs tokens), so do not compare them directly. Practical advice: spell out the subject, action, camera movement and lighting in the prompt; Flash has no keyframes, so camera changes must be driven by the prompt itself — do not expect keyframes to add control later.

Text-to-video: kling.ai/document-api/api/video/4-0-flash/text-to-video Image-to-video: official docs carry a separate image-to-video path (single image = first frame) Omni: up to 15 inputs (≤10 images + ≤5 videos) Billing: no input video 0.6 Units/s · with input video 0.4 Units/s (2x billing) Caps: 3–20 s · 720p · no keyframes · 8-bit SDR

Route 3: MCP server and CLI (new Nov 4)

Opened alongside the Flash GA: an MCP server and a command-line tool. MCP fits agent wiring: an agent can call Flash to generate samples and hand the results downstream; CLI fits batch scripts: dozens of draft clips across a prompt matrix in one run. Details per Kling's docs; this site has not yet verified the command parameters one by one — read the official docs before writing call code, and do not rely on this article's summary.

Route 4: resellers and AI plugins (mind the sourcing)

Resellers: UnificAlly lists it (kuaishou/kling-4-0-flash-video, from $0.0756/s, Nov 5); Runway also carries Flash (per ReviewRamble). This site has not verified one by one whether resellers apply the same 2× rule, so check the reseller's own page before ordering.

AI plugins: Adobe Photoshop / Premiere Pro / After Effects — per Kling's X post as reported by secondary sources, not verified against an Adobe listing, so this site marks them second-hand. Do not plan an editing workflow on "already available in Adobe" — confirm the plugin status through Adobe's official channels first. Kling 4.0 separately entered Adobe Firefly on Nov 10 (per ReviewRamble) — that is the full 4.0, not Flash.

The Omni reference workflow: how to use 15 inputs

Flash's Omni takes up to 15 inputs (≤10 images + ≤5 videos). A suggested workflow in three steps: first, fix the subject — use image references to lock the look of a person, product or scene; the full model supports up to 7 subjects, while the Flash tier's subject count is not separately published, so validate multi-subject consistency at small scale before batching. Second, fix the motion — use video references to set camera movement and action; remember that input video triggers 2× billing, so more references mean higher cost — do the math before batching. Third, fix the first frame — with single-image image-to-video, that image is the first frame, and its composition sets the clip's starting point; Flash has no last frame, so the ending is uncontrolled — leave the closing shot to the full model.

The 15-input cap is a count, not a quality guarantee: 15 inputs do not all take effect equally, and when references conflict, the official docs' priority rules apply (this site has not verified the priority rules one by one). In practice, fewer is better: 2–3 strong references are usually more stable than 10 weak ones.

The trade-off: Flash for drafts, the full model for finals

ScenarioPick FlashPick the full Kling 4.0
Batching samples, validating ideasYes: $0.094/s, 10s ≈ $0.94No: ~67% more at the same 720p
Fast plain-text clipsYes: the no-input-video row is cheapestNo
Keyframe storytellingNo: no keyframes, no last frameYes: up to 10 keyframes
1080p/4K deliveryNo: 720p ceilingYes: 1080p shipping, 4K coming soon
30s clipsNo: per-request cap 20sYes: 3–30 s
Input video / Omni multi-referenceUsable, but do the 2× math first (effective ≈ $0.126/s)Usable; the with-input-video billing row was not separately listed in this round — check the official docs
Verified native audioNo: audio unverifiedYes: stereo on every clip (per Genfire, citing official wording)

The division of labour is simple arithmetic: batch drafts on Flash and pick a direction; then hand the chosen prompts and references to the full model for keyframes and 1080p finishing. Do not do it backwards: do not build a keyframe flow on Flash (it has none), or generate 30s clips on Flash (it cannot), and then migrate to the full model — the migration costs more than splitting the tiers from the start.

FAQ

What are Flash's API paths?

Three official API paths: text-to-video, image-to-video, Omni multi-reference, published with the Nov 4, 2026 GA. Docs: official API docs.

How do I connect MCP and CLI?

Opened with the Nov 4 GA: the MCP server fits agent wiring, the CLI fits batch scripts. Details per Kling's docs; this site has not yet verified the command parameters one by one.

Are the Adobe plugins usable yet?

Unverified. The Adobe Photoshop / Premiere Pro / After Effects plugins rest only on Kling's X post via secondary sources, not verified against an Adobe listing; do not plan an editing workflow on "already available" — confirm through Adobe's official channels first.

How many references does Omni take?

Up to 15 inputs (≤10 images + ≤5 videos). The full model additionally supports up to 7 subjects and audio/voice references; the Flash tier's subject count is not separately published, so validate multi-subject scenes at small scale first.

What is Flash good for — and not good for?

Good for: batching drafts, validating ideas fast, plain-text samples. Not good for: keyframe storytelling, 1080p/4K delivery, 30s clips, verified native audio — use the full Kling 4.0 for those.

ModelBrief's take: Flash has one rule of use: drafts on Flash, finals on the full model. The API, MCP and CLI are all new channels opened Nov 4, and the integration cost is low; the real costs sit in the two unverified items — the Adobe plugins are not confirmed against an official Adobe listing, and the audio claims conflict. Run a few clips on the website first, confirm the style and the caps, and only then write Flash into your workflow.

Reference: Kling official API docs: 4.0 Flash text-to-video · Kling official usage guide: Kling 4.0 (§5.1) · Kling official pricing page · OrcaRouter Flash GA analysis (Nov 4, 2026, updated Nov 7) · UnificAlly (Nov 5, 2026) · ReviewRamble (Nov 14, 2026) (rewritten and organized by ModelBrief; unverified items are marked)