Guide
English
Search the docs⌘K
3.5

Reworking the picture

Same character, same scene — in an AI video you often need another take with a different expression, a different camera angle or different light. DOTV's retouching tools are a reshoot: Expression tuning, Multi-angle, Relight, 720 panorama and Split grid — the person stays the same person and the set stays the same set; only the one thing you asked for changes. Except for Split grid, each one drops a new node to the right of the original and wires the original in as a reference — the original itself never changes.

Expression tuning

Same person, same camera, same clothes — only the expression changes. Open the Reshoot dropdown on the toolbar and pick Expression tuning.

First tell it who is who: it runs face detection, draws a box on each face it finds, and clicking a box picks that person. Faces in profile and people in the background often go undetected — the hint bar along the top of the image has Draw a box, so drag one around them yourself. If there are several people in the frame you can box several of them and tune each expression on its own.

On the left a plain white model head, on the right a 5×5 pad holding the 25 expressions you can pick from.
  • The two axes of the pad: Close ↔ Distant across, Excited ↔ Calm down, 5×5 for 25 cells in all, with "Neutral" dead center.
  • Hold the ball on the pad and drag it around; let go and it snaps to the nearest cell. The white head on the left changes with it, so you can check this is the one you meant.
  • Every cell has a name ("Ecstatic", "Annoyed", "Severe" and so on), shown on the Mood row.
  • With several characters, pick one at the top before you choose. Switch to someone else and back, and the cell you set is still there.

Multi-angle

Same scene, shot again from another camera position. Open Multi-angle in the Reshoot dropdown and a camera editor fills the strip under the image.

Presets along the top; below them the camera sphere on the left and sliders for the same parameters on the right.

The three ways of placing the camera are three ways of saying the same thing — move one and the other two follow:

  • Presets — Fisheye, Dutch angle, High angle, Low angle, Bird's eye, Rear view. One click sets all three parameters at once.
  • Camera sphere — a camera hangs on a wireframe sphere; hold it and drag it around the subject. Drag sideways to orbit, drag up and down to tilt. The thumbnail at the middle of the sphere is the thing being shot.
  • Sliders — Orbit, Tilt and Shot scale (Close-up / Close / Medium / Full / Wide). Use these when you need one exact angle.

Move any of them and the preset row falls back to Custom. Prompt is collapsed by default; open it when you want to add something on the side ("and move the light round to one side"). Reset next to it clears only the camera and the preset — what you typed stays.

Relight

Nothing in the frame moves; only the lighting is done again. Open Relight in the Reshoot dropdown and the panel has three columns: place the lights on the left, tune the current light in the middle, say what mood you want on the right.

Light sphere on the left, key light and rim light in the middle, smart mode on the right.
  • Light sphere — drag the light around the subject. Wherever you want the light to come from, that is where you put it.
  • Key light — six common positions: Left, Top, Right, Front, Bottom, Back.
  • Brightness — overall exposure. 50% keeps the original, drag left to darken, right to brighten. Color tints the key light; clear it and the original light color is kept.
  • Rim light — turn it on to add a second light behind the subject, at one of nine angles in a 3×3 grid.
  • Smart mode — the column on the right (it has a switch; turn it off and the whole column folds away): eight presets (Blown-out film, Blue backlight, Rembrandt lighting, Cyberpunk, Sunset haze, Moody dark, Golden hour, Nolan gray), a one-line description, and a reference image. The reference image is copied for its light, not its content.

720 panorama

Stretch an ordinary frame into a panorama you can turn around inside: 360° across, 180° up and down. Pick 720° panorama from the Preset panel on the toolbar and a pending node lands to the right of the original.

The image node toolbarIt only shows when you select an image node that has already generated. Everything in this chapter starts from this bar.
The 720° panorama entry in the toolbar's Preset panel: build a 720° panorama from this scene.

Two things on this route are fixed for you and cannot be changed in the composer: the aspect ratio is locked to 2:1 (that is simply what a panorama is — at 16:9 the image stretches top to bottom in the viewer and the two ends do not meet) and the model is locked to GPT Image 2.5 (nothing else gives you a true 2:1 canvas).

Once the image is out, a Panorama entry appears in the top right corner of the node. Click it and the node starts turning — the bar above the image is replaced by the panorama viewer's own:

Exit and fullscreen at the two ends; the five in the middle are three kinds of capture, reset view and composition guides.
  • Drag with the mouse to change where you are looking, scroll to change the field of view. On the keyboard, Q / E dolly in and out, W / S tilt up and down, A / D pan left and right.
  • The toolbar also has Reset view and Composition guides, plus Fullscreen preview (your angle carries over when you come back out).
  • Three captures: Capture this view, Capture 4 views (four around the circle) and Capture 12 views (twelve around the circle). What comes out drops straight back onto the canvas already wired up, as ordinary image nodes.

Split grid

The nine-cell and twenty-five-cell sheets a preset gives you are one image, not nine. Split grid cuts the cells you want into separate images.

Click Split grid on the toolbar and pick a size first. Four common ones: 4 cells (2×2), 9 cells (3×3), 16 cells (4×4), 25 cells (5×5). What the model returns is not always a square block (a "2 rows by 4 columns" strip cut as 2×2 comes out completely misaligned), so for those go to Custom grid and drag the rows and columns out on the small grid (up to 5×5).

The grid lies over the image; click a cell to select it, and the bar along the bottom tells you how many you have.
  • Click a cell to select or deselect it. Hold Shift and click a second one to take the whole rectangle between the two.
  • The bar along the bottom reads N cells selected. With nothing selected, the two actions on the right are grayed out.
  • New shot group — each selected cell is cut into an image of its own, and they all go into a new group on the canvas. Each is named for where it sat in the grid ("1-2" = row 1, column 2).
  • Upscale image — cuts them out first, then upscales cell by cell. Pick the factor on the button to the right (2× or 4×).
Try it in DOTVReading only goes so far: open DOTV and make an AI video with the feature this section covers.