RunwayImage-to-videoWorkflow & organisation

Runway Aleph: Edit Video With Text Instructions

By Marco Cavazzana · Updated · 5 min read
ShareEmail
Short answer

Runway Aleph is an in-context video model: you upload an existing clip and type an instruction such as "make it night", "remove the car" or "show this from a low angle", and it returns a new version of the clip with that change applied while keeping everything else. It edits what is already there rather than generating from scratch.

Most AI video tools make new footage. Runway’s Aleph, released in 2025, changes footage you already have. You give it a clip and a sentence, and it returns the same clip altered: a different object in the actor’s hand, a scene relit for dusk, a new camera angle on the same action, a distracting sign removed, or the shot that would logically come next. This guide explains what kinds of instruction Aleph handles well, how to phrase them and how to organise a project so edits do not pile up into confusion.

What Aleph can change

Aleph is described by Runway as an in-context model, which means it reads the whole clip and makes the edit in relation to it, rather than treating your instruction as a fresh prompt. The main categories of edit are:

  • Object changes. Replace, add or swap something in the scene: turn a coffee cup into a glass of water, put a hat on a character, change the colour of a car.
  • Relighting. Change time of day, weather or light quality: golden hour, overcast, neon night, candlelight.
  • Camera angle. Ask for the same moment from a different position: a low angle, a wide shot, a view from behind the subject.
  • Removal. Take out people, logos, cables, text or clutter and let the model fill the gap.
  • Next shot. Generate a plausible following shot that continues the scene, useful for coverage you forgot to film.

What it does not do is turn a clip into a completely different scene. The more your instruction preserves the original, the more reliable the result. Aleph is a finishing and fixing tool first, and a generation tool second.

Preparing the source clip

Aleph works from your footage, so footage quality sets the ceiling. Use a clip that is sharp, reasonably stable and short; a few seconds is enough for most edits and keeps credit costs down, since Runway bills per second of generated video. Trim the clip to the exact range you need before uploading, because every second you include is a second the model processes and you pay for. Check Runway’s current pricing page for Aleph costs.

Clips with a clear main subject and a simple background are the easiest to edit. Heavy motion blur, flickering lights and fast whip pans make it harder for the model to track what should stay and what should change. If the clip came from Runway’s own image-to-video (see the Gen-4 guide), Aleph is a natural second pass to fix a detail you could not control in generation.

Writing an edit instruction

The instruction is an edit note, not a scene description. The phrasing that works best is short, specific and names both what to change and what to keep.

  1. One change per instruction. "Make it night" or "remove the bicycle", not both at once. Stack edits in separate passes so you can judge each one.
  2. Name the target precisely. "The red bicycle leaning on the wall on the left" beats "the bike" when there are several objects in frame.
  3. State what stays. "Keep the actor, the framing and the camera movement the same." This stops the model from helpfully improving things you did not mention.
  4. For relighting, describe the light, not the time. "Warm low sun from the right, long shadows" is more controllable than "sunset".
  5. For camera angles, say where the camera goes. "Same action from a low angle near the floor, looking up" rather than "make it more dramatic".
  6. For next shots, describe the continuity. "The next shot: she walks through the door we see in the background, interior, same lighting."

You can also supply a reference image alongside the instruction in some modes, for example to show the exact object or style you want introduced. Where available, this is more reliable than describing a specific product or texture in words.

Settings at a glance

SettingWhat it doesStart with
Source videoThe clip to be edited; everything not mentioned is preservedA trimmed clip of a few seconds, sharp and stable
InstructionThe text edit noteOne change, named target, what to keep
Reference imageOptional visual example of the object or look to introduceUse when words are ambiguous
Aspect ratioOutput shape; normally follows the sourceMatch the source
Fixed seedRepeats the same randomness so instruction changes are comparableOn when refining a stubborn edit

Labels and the exact options in the Aleph panel have shifted since launch, so look for the equivalents rather than these words.

Keeping an edit chain organised

Aleph invites you to edit an edit, and three passes later nobody remembers what the original looked like. A simple system prevents that:

  • Name exports in sequence with the instruction in the filename: shot04_v1_source.mp4, shot04_v2_night.mp4, shot04_v3_remove-sign.mp4.
  • Keep a short log per shot with the instruction, the model, the seed and whether the pass was accepted.
  • Always edit from the best accepted version, not the latest attempt, so a failed experiment does not become the base for everything after.
  • Decide early whether colour grading happens in Aleph or in your editor. Doing it in both leads to double-graded footage.

For a broader approach to keeping records of client work across tools, see keeping a record of AI work for clients.

Try it

Change the scene to late evening: warm low sun from the right, long soft shadows, sky turning orange. Keep the actor, her movement, the framing and the camera motion exactly the same.

Swap the lighting description for whatever your scene needs, or replace the first sentence with a removal ("remove the parked white van on the left") or an angle change ("show the same action from a low angle"). Always keep the final sentence so the model knows what is off limits.

Frequently asked questions

What is Runway Aleph used for?
Aleph edits existing video from a text instruction. You can change or remove objects, relight the scene, view the same action from a new camera angle, or generate a plausible next shot, all while keeping the rest of the clip as it was.
Is Aleph the same as Runway Gen-4?
No. Gen-4 generates new video from a text prompt or a start image. Aleph takes a video you already have and modifies it according to an instruction, so the two are used at different stages: Gen-4 to create, Aleph to fix and finish.
Why did Aleph change things I did not ask it to change?
The instruction probably left room for interpretation. Make one change per pass, name the target precisely and add a sentence stating what must stay the same, such as the actor, framing and camera movement.

Nothing you makeis lost.

Free · No card · Read-only · 2-minute setup

DreamdriveFree plan · Chrome
Add to Chrome — it’s free