Chain prompts — outputs feeding the next prompt

What chain mode does, in one paragraph

Normal runs treat each prompt as independent — prompt 1’s output has no influence on prompt 2’s input. With Chain prompts turned on, every prompt’s output becomes the reference input for the next prompt. A list of 6 prompts becomes a 6-step ladder where each step builds on the last. The result is a coherent sequence instead of 6 disconnected generations.

Where to find the toggle

Below the prompts textarea (and below the auto-match controls when those are visible), there’s a card with one checkbox: Chain prompts.

“Use each prompt’s output as the input for the next.”

When chain mode is on and you have 2+ prompts, a tiny arrow diagram appears below the checkbox showing the chain:

#0 → #1 → #2 → #3 → #4 → #5

Not available in every mode

Chain prompts is hidden in Text to Image. There’s nothing to chain — every image is a finished output with no handoff back into the composer. Available in:

• Text to Video — first clip’s last frame becomes the second clip’s start frame.
• Image to Image — first image becomes the second prompt’s reference image.
• Image to Video — first clip’s last frame becomes the second clip’s start frame.
• Reference to Video — first clip’s last frame becomes the chained start, overriding any @reference images in subsequent prompts. Use with care.

A worked example: 4-shot Image-to-Video sequence

Goal: a short cinematic sequence starting from a single still — a red panda on a mountain, gradually zooming out to reveal the landscape.

1. Mode: Image to Video.
2. Frame mode: Start frame.
3. Reference library: one image, 01-redpanda-closeup.jpg.
4. Chain prompts: on.
5. Prompts (4 shots, blank-line separated):

Slow push-in on the red panda's face, eyes catching the light. Subtle breathing motion.

Smooth pull-back, the red panda now full-body, sitting on a snow-dusted boulder.

Continue pulling back, revealing the mountain peak the panda sits on, dawn light.

Wide aerial reveal, the whole mountain range stretches into the distance, golden hour.

Click Run. The 4 clips chain together: shot 1 ends, its last frame becomes shot 2’s start, and so on. Stitch them in any editor and you get a single ~24s continuous sequence from one starting image.

A worked example: Image-to-Image style ladder

Goal: take a logo and step it through 6 progressively more painterly styles, ending in oil-painting.

1. Mode: Image to Image.
2. Reference library: one image, logo.png.
3. Auto-match: off, Max input: 1.
4. Chain prompts: on.
5. Prompts:

Same logo, slightly softer edges, watercolor wash background.

Increased painterliness, visible brush strokes, muted palette.

Strong impressionist treatment, broken color.

Oil-paint style, thick impasto, dramatic lighting.

Renaissance-style chiaroscuro, gold leaf accents.

Final: museum-quality oil portrait of the logo, signed corner, varnished.

Each step uses the previous output as its source, so the logo evolves through a continuous stylistic ladder rather than 6 random reinterpretations.

Things to know before you commit

• Chain runs are slow. Because the next prompt can’t start until the previous finishes (and the output is wired back in), there’s no parallelism. A 6-prompt chain takes ~6× as long as a 6-prompt independent batch.
• Failures break the chain. If prompt 3 fails after all retries, prompts 4–6 are marked skipped (you’ll see a skipped badge) because they have nothing to chain from. Re-running with Continue picks up from the failure.
• Chain is per-mode. Each generation mode remembers its own chain-on/off state separately, so turning it on in Text-to-Video doesn’t enable it in Image-to-Video.
• Don’t combine with Reference-to-Video @references carelessly. From prompt 2 onward, the chained start frame replaces whatever your @reference would have provided. The first prompt still uses @reference normally; later prompts effectively become Image-to-Video-style continuations.

When not to chain

Independent prompts beat chained ones for most use cases — A/B variants, thumbnail batches, style explorations on the same source. Save chain mode for the two situations it’s actually built for: continuous video sequences and stylistic evolution ladders.