Edit mode changes an existing clip rather than generating a new one. Because it conditions on real footage, the framing, lighting and identity in your source carry forward instead of being re-guessed. What it doesn't do is target an area — there's no mask and no region selector, so you describe the change and the whole clip regenerates against it. Nine models support it.
The expensive part of AI production isn't generation. It's starting over on something that was mostly right.
What does Edit mode actually do?
Regenerates a clip against a description, using the clip as the starting condition.
The standard failure looks like this. You generate a shot. The performance lands, the light is right, the framing is what you wanted — and one element is wrong. Regenerate from scratch and the new take fixes that element and loses the performance.
Edit mode changes what you're regenerating from. Instead of a description and nothing, it's a description and a clip. The framing, the light, the identity and the space are all present in the source, so they persist without being described.
A new shot starts from a description. An edited shot starts from a shot.
That's the whole difference, and it's worth being precise about what it isn't. You aren't preserving part of the frame. The entire clip comes back regenerated. What changes is that the model had something to work from.
Why isn't there area selection?
Because there isn't, in either mode, and it's worth knowing before you plan around it.
Video Edit takes a model, a mandatory source clip of three to ten seconds, references, a description of the change, quality, and a sound toggle where the model has one. No mask. No brush. No box.
Image Edit is described in Hexcoded's own documentation as image-to-image editing and targeted inpainting — but targeting happens through the prompt rather than through a selection. It requires a mandatory base image, and it restricts references to visual ones: image and actor references work, while video and audio reference tabs are disabled in that mode.
So in both cases, precision comes from a precise description rather than from pointing at something.
What is it good at, and where does it fall short?
Three things, and the limits matter as much as the strengths.
| What Edit mode is good at | Why | Where it falls short |
|---|---|---|
| Keeping what worked | The source clip conditions the generation, so framing, light and identity persist without being described | The whole clip still regenerates. Nothing is preserved pixel for pixel |
| Changing a described element | A prompt against real footage is a far narrower instruction than a prompt against nothing | You can't point at the thing. If the description is ambiguous, so is the result |
| Restyling a clip | Wan 2.6 exists specifically for this — the picker describes it as "restyle a clip you already have" | A restyle changes everything by definition, which is the point and also the limit |
Source: Hexcoded Creative Studio documentation and picker, checked September 2026.
Where this falls short. There's no boundary to work inside. Everything comes back regenerated, so a change you asked for can arrive alongside changes you didn't. The source constrains the result; it doesn't lock any part of it.
What can't it fix?
Three things, and each fails for a different structural reason.
Identity drift. Identity is carried across the whole frame and the whole duration, so a described change to a face tends to alter more than you asked for. Drift is a reference problem — a saved actor called by handle is the fix, not an edit.
Lighting. Light is global by nature. Describing a change to one part of the scene's light leaves the rest lit from somewhere else, because the whole frame regenerates against a description that only covered part of it.
Performance. Timing, gesture and expression are properties of the whole take. If the read is wrong, regenerate — there's nothing narrower to change.
What are the constraints worth knowing before you generate?
Four, and the first will catch you out if you don't plan for it.
A three-to-ten-second source clip. That's a hard range. A thirty-second single take from Seedance 2.5 can't be edited as one piece — it's outside the window entirely. If you intend to fix a long take later, that intention doesn't survive contact with Edit mode.
Aspect ratio is inherited. In Edit, Extend and Motion, the aspect ratio comes from the source clip and the selector is hidden. Framing is decided in the first generation.
Nine models, not all of them. Edit mode is supported by Seedance 2.5, Seedance 2.0, Seedance 2 Fast, Seedance 2 Mini, Gemini Omni Flash, Grok Imagine 1.5, Wan 2.7, HappyHorse 1.0 and Kling 3.0 Omni. The dropdown filters automatically to whichever models support the mode you're in.
Image editing has a smaller reference pool. Sixteen assets against video's fifty, and video and audio references are disabled in Edit mode.
How does this change your review process?
It moves review earlier and makes it more specific.
Without Edit mode, reviewing a shot is close to binary — keep or regenerate — so there's little value in cataloguing exactly what's wrong. With it, the useful review output is a description of the change you want, because that description is the input.
The practical change: review with a note-taking habit rather than a verdict habit. "The mug should be on the left of the table, and the light should stay as it is" is an actionable instruction. "It's not quite right" isn't.
How should you actually use it?
Four steps, and the order saves the most time.
Review the assembled sequence before the shot
Assemble first, then decide which shots survive. Editing a shot you'll cut is the most common wasted effort.
Check the clip length
Three to ten seconds. If your take is longer, Edit mode isn't available to you and regeneration is the only route.
Sort the problems into fixable and not
A described element goes to Edit. Identity, lighting and performance go to regeneration. Sorting first stops you attempting changes that can't work.
Describe the change and what should stay
The whole clip regenerates, so saying what should remain is as useful as saying what should change. Attach the same references you used originally.
- Edit mode regenerates a clip against a description, using the clip as the starting condition. Framing, light and identity carry forward
- There is no area selection anywhere — video or image. Precision comes from a precise description
- The whole clip comes back. Nothing is preserved pixel for pixel
- Source clips must be three to ten seconds. A thirty-second take can't be edited as one piece
- Aspect ratio is inherited and the selector is hidden. Framing is decided in the first generation
- Nine models support Edit mode. The dropdown filters to whichever ones do
- Identity, lighting and performance can't be fixed this way. All three are properties of the whole take
- Review the assembled sequence first. Editing a shot you'll cut is pure waste
It changes an existing clip rather than generating a new one. Because the generation is conditioned on real footage, the framing, lighting and identity in your source carry forward without being re-described.
No. There's no mask and no region selector in video or image editing. You describe the change and the whole clip or image regenerates against the source. Targeted results come from a precise description rather than a selection.
Between three and ten seconds. That's a hard range, so a thirty-second single take can't be edited as one piece — which matters if you generated it intending to change something later.
Not reliably. Identity is carried across the whole frame and the whole duration, so a described change to a face tends to alter more than you asked for. Drift is a reference problem — a saved actor called by handle is the fix.
Nine: Seedance 2.5, Seedance 2.0, Seedance 2 Fast, Seedance 2 Mini, Gemini Omni Flash, Grok Imagine 1.5, Wan 2.7, HappyHorse 1.0 and Kling 3.0 Omni. The model dropdown filters automatically to whichever models support your current mode.
When the problem is identity, lighting or performance — all three are properties of the whole take. Regenerating with a corrected description, or with a saved reference attached, is faster than describing your way out of it.
Start from a shot, not from nothing
Edit and Extend sit alongside Create in Creative Studio, on 30+ models and one credit balance. A shot that nearly worked becomes an input rather than a write-off.
Open Creative StudioMore on model capability, access and rights in Models.