New camera angles without a reshoot means what it says: one clip you already have, viewed from a close-up, a profile, an over-the-shoulder, or a high angle — while the performance stays the same second for second. You are not directing a new scene. You are moving the camera around a scene that already exists.
The official NeuroHelper method is a conversation in Chat Master plus a generate step in Video Master with Seedance 2.5 Video Edit. Chat Master studies your source clip, suggests a short camera plan from a shot-and-angle library, maps each viewpoint to the original timing, and writes one English prompt. You paste that prompt with the same source video and generate. The long-form walkthrough below is the official film to watch; a related short exists, but this is the one to learn from. The source PDF is almost typographic, so the video is the only large media on this page.
The workflow: Start with a 4–30 second, well-lit medium shot you have the right to edit → attach it in Chat Master and ask for a camera plan (or let it propose 3–4 options) → confirm the sequence and timing → copy the finished Seedance prompt → in Video Master, choose Seedance 2.5 Video Edit, attach the same source video, paste, generate → review that the performance was recamera’d, not “improved.”
The source clip is the performance master
That sentence is the whole method. The source video controls identity, face, hair, body proportions, expression, lip movement, gestures, timing, rhythm, and emotion. You may change shot size, camera angle, camera position, perspective, a camera move, or — if you explicitly ask and attach a reference — a location or an object. You do not invent a new action, a new feeling, a new outfit, or a new tempo unless you mean to. At a cut, the person must be in the same phase of movement as the original at that exact second. A close-up is a new view of the same moment, not a chance to restart a gesture. The same habit shows up in the AI realism playbook and, in a different way, in character replacement — one element changes, the performance does not.
Prepare the right source video
Use a clear, high-quality clip with even lighting, 4–30 seconds long. Shorter or longer than that range is not a good fit for this workflow. A medium shot is usually the safest start: the face should not be extremely close, and it should not be a small figure in a wide landscape. Hands and body should be visible when they matter. Faces should not sit in hard one-sided shadow. Use footage you own or have the right to recamera.
Plan the cameras in Chat Master, then generate
Attach the clip in Chat Master and describe what you want in plain language — “give me a close-up and an over-the-shoulder,” or “choose a short plan for me.” You do not need filmmaking vocabulary. Chat Master should watch the whole clip first, name what actually happens, and only then suggest viewpoints that support that action. If you are unsure, ask for three or four suitable options, each explained in a sentence. Natural cinematic choices should come first. Experimental looks such as fisheye, a Dutch angle, a prism, or an inside-the-fridge shot should not be the default unless they genuinely suit the scene or you ask for something more playful. A short, simple clip may need only two or three strong viewpoints. Do not force a change every few seconds without a visual reason.
Once you confirm the shots, Chat Master should map each one to the source timing and return a single English Seedance prompt. In Video Master, choose Seedance 2.5 Video Edit, attach the same source video, paste that prompt in full, and generate. Confirm quality and output in the picker (the official instruction is written for this Video Edit path). If something drifts, do not rewrite the entire brief — strengthen only the failing lock: identity, motion phase, or a single camera block.
A practical shot-and-angle library
These are the official starting points, written in plain language. Use them as a menu, not as a checklist to exhaust in one clip.
Shot sizes
- Wide — the subject plus more of the surroundings.
- Medium — a conversational distance, gestures still readable.
- Close-up — the face or an important detail, closer to the viewer.
- Macro — an extremely close detail of a small object or texture.
Natural angles
- Profile, three-quarter, rear — side view; a view between front and profile (about 45 degrees); the camera behind the subject.
- Over-the-shoulder — looking past the subject’s shoulder toward the action.
- POV and eye-level — the scene from the subject’s view, or a camera at about eye height for a neutral stance.
Height and tilt
- Low-angle / high-angle — looking up from below, or down from above.
- Overhead / aerial — straight down from directly above, or a much higher top view that shows more environment.
- Ground-level / worm’s-eye — very close to the floor, or an extreme upward view.
- Dutch angle — a deliberate tilt so the horizon is diagonal. Use it only when the scene can carry the unease.
Unusual viewpoints and lens looks (when they actually fit)
Framed or playful options in the official library include looking out from inside a hole or cavity, a hoop-level camera, a selfie-style handheld feel, and an inside-the-fridge view. Lens variations include a wide-angle close-up, fisheye, fisheye close-up, high- or low-angle close-ups, and an overhead fisheye. Optical effects include a prism reflection, a silhouette, foreground occlusion, a focus swap between foreground and background, a deliberate flare, or a magnifying-glass look. Default to the natural set first. Gimmicks are available; they are not the method.
How the finished prompt should be structured
Ask Chat Master to return one clean English prompt, not a conversation transcript, using this spine:
BASE
Use @video1 as the base video.
SOURCE ACTION
[A short description of what actually happens in @video1]
PRESERVE
[Everything that must stay unchanged: identity, wardrobe, emotion, timing, environment unless you asked to change it]
EDIT
[Only the requested camera or environment changes]
CAMERA AND SHOT BLOCKS
[For each time block: the new camera setup, plus an explicit reminder to keep the source action at those exact seconds]
Example logic:
0–4s — source action continues unchanged, medium eye-level view
4–8s — same movement and same motion phase, close-up three-quarter view
8–12s — same continuing action, over-the-shoulder view
REFERENCES
[If you attached stills: @image1 = environment, @image2 = object, and so on. Each still gets one job.]
MOTION CONTINUITY
Preserve the original action and exact motion phase throughout the edit.
CONSTRAINTS
No face drift. No identity change. No random outfit changes. No anatomy errors. No broken hands. No extra limbs. No additional people unless requested. No unnatural motion. No flickering. Keep the result realistic and visually continuous.
A camera cut changes only how the action is viewed. It must not reset a body position, repeat a gesture, add a movement, or change the timing of an object interaction. If Chat Master is missing information, it should ask rather than guess.
How to do this in NeuroHelper AI
- Chat Master. Attach a 4–30 second source clip you have rights to use. Ask for a short plan, or for a complete sequence if you want it chosen for you.
- Video Master. Seedance 2.5 Video Edit, the same source video, the finished prompt. Confirm output settings in the picker.
- Image Master, only if you are also replacing a location or an object. Give each still one explicit role. The source video still owns the performance. A current still model such as GPT Image 2.6 is enough for those references.
- Audio Master. Only if you need to clean or replace a bed in Audio Master after the picture is right. Do not let a new score disguise a broken motion phase.
Compare video tools in the models hub, best AI video generators, and Seedance vs Kling. Related craft: cinematic action, bullet time, a drone path from a map, and a reusable character sheet if the face will recur. A related short on the same idea is on the official channel (jrK48N5HlLM); the film embedded above is the long-form walkthrough to follow.
Mistakes that secretly rewrite the performance
- Choosing angles because they look “cinematic” in isolation. They have to support what is already happening.
- A camera change every two seconds. Fewer stronger viewpoints usually read more cleanly.
- Defaulting to fisheye, Dutch tilt, or inside-the-fridge. Those are library options, not the starting point.
- A cut that restarts a gesture. The motion phase at 0:08 must match the source at 0:08.
- Letting a location still overwrite the performer’s identity. Assign each reference one job; the video remains the performance master.
Watch the long-form recamera, then try one clip
The test is recognition and timing: it should feel like the same performance, photographed from seats you did not have on the day. Watch the official film, then attach one 4–30 second medium shot you have the right to edit.
Frequently asked questions
Can AI really create new camera angles without a reshoot?
In this workflow, yes: you start with a 4–30 second source clip, plan a short sequence of new viewpoints in Chat Master, and generate the recamera’d version in Video Master with Seedance 2.5 Video Edit. The source performance stays the master. Each new shot is a different view of the same moment, not a rewritten scene.
How long should the source video be?
The official instruction is 4–30 seconds. Clips shorter than 4 seconds or longer than 30 seconds are not a good fit. A clear, evenly lit medium shot is usually the safest starting point.
Which model generates the new angles?
The official path is to write a Seedance prompt in Chat Master, then generate in Video Master using Seedance 2.5 Video Edit with the same source video attached. Confirm quality and output in the current picker before you generate.
Should I start with unusual angles like fisheye or inside-the-fridge?
Not by default. The official library includes those options, but the method is to prioritize natural cinematic choices that fit the existing action, and to use experimental viewpoints only when they genuinely suit the scene or when you ask for something more playful.
Is there a shorter YouTube version of this walkthrough?
Yes. The related official short is jrK48N5HlLM. This page embeds the long-form film szSQaeKzBTk (about five minutes), which is the better teaching version.
How much does a recamera pass cost in NeuroHelper AI?
It depends on the model, duration, resolution, and references you choose. Video Master shows the current estimate before you generate. You can start with the $1 trial.