Cine con IA · Director de camara H3

The single highest-leverage phrase in a video prompt is the one that says where the camera is. Naming a real lens or a real shot size moves a model out of its default framing faster than any pile of "cinematic, 8k, masterpiece" ever will - the encoder was trained on captions written by people who owned cameras. The problem is that most of us write that line badly, in the wrong language, and mix it into the middle of a paragraph about what the character is wearing.

This node is the fix. It's the camera department, split off from the script.

What it actually does

You pick a shot size, an angle, a movement, an intensity, a focal length and a depth of field from Spanish dropdowns, and it compiles them into English camera sentences in the same structure MiniMax H3 was prompted with. Then it stitches that clause into a finished six-section H3 prompt.

The mappings are literal, not vibes. plano medio becomes "a medium shot". contrapicado becomes "with the camera below the subject, looking up". 85 mm becomes "The image uses a 85 mm lens." Put a shot size and an angle together and they share one sentence; leave one of them on sin especificar and the other one carries the sentence alone. Two details worth knowing because they'll confuse you otherwise: intensity only modifies movement, so marcada does absolutely nothing unless you've also picked a movement, and the depth-of-field options are real photographic language ("The shot has very shallow depth of field"), not the word "bokeh" scrawled in a corner.

The instruccion_camara field is the escape hatch

Everything above is a template. instruccion_camara is a free-text box for a full trajectory or a complete camera recipe - the sort of verified camera clause people publish for H3 - and the node preserves it verbatim as its own paragraph, then appends the structured clause after it. That ordering is deliberate: it means a recipe you pasted in isn't silently rewritten by whatever dropdown position you left behind.

reglas_continuidad is a boolean, on by default, that appends a short constraints block to the end of the prompt: identity, anatomy, clothing and object contact stay consistent, nobody wanders into frame who wasn't established, and the requested camera path wins over everything else. It only appends once, even if you re-run.

The node needs a scene to work on, and it will only take one from this pack - feed it the scene output of Cine con IA · Prompt simple H3 or of the eight-box Escena por apartados H3, not a naked string from a text node. Wire the wrong type in and it stops with "Camera Director necesita la salida scene de CineConIA Scene / Prompt H3".

Inputs and outputs that matter

Required: scene (the CINECONIA_H3_SCENE wire), plano, angulo, movimiento, intensidad, lente, profundidad_campo, instruccion_camara, reglas_continuidad. Every dropdown defaults to sin especificar, which contributes an empty string - so a fresh node adds nothing and you can add one control at a time.

Two outputs. prompt (STRING) is the finished text; it goes into the Escena node of the same pack, which turns an H3 prompt plus your reference images into conditioning and the audiovisual latent. info (STRING) is a one-line summary naming the camera text it built or telling you there's no camera instruction at all - drop it into a text preview node and you can see what the node decided without rendering.

Install

Same pack, same install, for all five H3 nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/chaLords/ComfyUI-Cine-con-IA.git

Restart ComfyUI and look for the Cine con IA category; the node itself is under Cine con IA/H3. The pack installs no Python libraries of its own and has no requirements.txt to fight with, which is genuinely rare and worth appreciating. It does want a recent ComfyUI with native MiniMax H3 support (the repo states 0.34.2+).

Two install notes specific to this pack. ComfyUI Manager may not find it: it's registered in the Comfy Registry as cine-con-ia, but at the time of writing the versions were still pending review, so search by the display name and you may get nothing back. And the interface is in Spanish - the node names, the dropdowns, the tooltips. If you found this page by googling "CineCameraDirectorH3", that's why nothing looked familiar.

Where people get burned

The most common one isn't a bug: people describe the camera inside the scene's action text and pick a movement here, and H3 gets two instructions that fight. The pack's own tooltip on the action field says it plainly - don't put camera movement in the action.

The second one is expecting a shot change. If your scene text already contains a [Shot 1] marker, the camera clause is injected right after it; if it doesn't, the node adds one. Either way you get one camera instruction, not a sequence. H3 honours the order of events in a prompt, not timestamps, so trying to schedule three moves across five seconds by listing them is not a thing this node can do for you.