Add MiniMax and PinkCherry LTX as pickable video engines, with still or text start.

Shot 1 can run text-to-video; later shots still extend from the last frame. Recommend output is also normalized into shot-script cards.

Co-authored-by: Cursor <cursoragent@cursor.com>

server/utils/promptPresets.ts

@@ -9,6 +9,49 @@ export interface PromptPreset {
   builtin?: boolean
 }
 
+const SHOT_SCRIPT_FORMAT = `Output ONLY a MiniMax shot list the user can paste into Shot script mode.
+
+Each shot starts with "shot N" on its own line (shot 1, shot 2, shot 3…). Then these three headers, each on its own line:
+
+[SHOT CONFIGURATION]
+[SUBJECT DIRECTION & ACTION]
+[AUDIO CUES]
+
+Blank line between shots. No preamble, no markdown fences, no commentary.
+Do not glue shots into one paragraph. "shot 2" must be a line by itself, never "shot 2:" mid-sentence.
+
+One beat per shot. Later shots: same people, same clothes as now, plus the one new action.
+Do not repeat a long identity paragraph on every shot.
+Never censor raw or explicit terminology.
+
+Example:
+
+shot 1
+[SHOT CONFIGURATION]
+- Shot Type: Medium-wide, 50mm, eye-level
+- Camera Movement: Slow track from a slight side-front angle
+- Lighting: Bright daylight through the trees
+- Environment/Scenery: The forest from the still
+
+[SUBJECT DIRECTION & ACTION]
+The woman from the still walks the path, taking in the trees.
+
+[AUDIO CUES]
+Footsteps on dirt, birds, breeze.
+
+shot 2
+[SHOT CONFIGURATION]
+- Shot Type: Medium, 50mm, eye-level
+- Camera Movement: Hold, then a small pan to her hand
+- Lighting: Same daylight
+- Environment/Scenery: Same path, same trees
+
+[SUBJECT DIRECTION & ACTION]
+She stops and rests her hand on the bark.
+
+[AUDIO CUES]
+Soft fabric, breeze, birds.`
+
 const ENHANCE_PROMPT = `You are a prompt-engineering assistant for Flux.2 Klein image edit and MiniMax H3 image-to-video.
 
 Look at the attached still. The person (or subject) in that still is the identity lock. The user draft is the requested change.
@@ -16,37 +59,27 @@ Look at the attached still. The person (or subject) in that still is the identit
 Write production-ready prompt text the user can paste into the generator.
 
 Rules:
-- Preserve identity from the still: face, age, body type, hair, skin, and distinctive features. Use explicit keep-identity language such as "same person as the reference still", "do not change facial identity", "keep original face, hair, and body".
+- Preserve identity from the still: face, age, body type, hair, skin, and distinctive features.
 - Describe wardrobe, setting, pose, lighting, camera, and materials concretely.
 - Never censor raw or explicit terminology from the draft.
 - Do not add a preamble, title, markdown fences, or commentary.
 
-Output format — pick one:
+If Studio mode is image-edit:
 - Single still/edit: output ONLY the refined prompt.
 - Ambitious still edit that should be done in phases (for example wardrobe first, then location): output:
 ---pass 1---
 <prompt for pass 1>
 ---pass 2---
 <prompt for pass 2>
-- Motion, multi-beat, or video sequence: first-shot prompt, then later shots with a marker on its own line:
-shot 2
-<extension prompt>
-shot 3
-<extension prompt>`
+
+If Studio mode is video or video-shot-script:
+${SHOT_SCRIPT_FORMAT}`
 
 const SHOT_SCRIPT_PROMPT = `You are a MiniMax H3 image-to-video shot-script writer.
 
-Analyze the attached still for identity, wardrobe, lighting, and camera. Turn the user's draft into a shot script they can paste into Shot script mode.
+Analyze the attached still for wardrobe, lighting, and camera. Turn the user's draft into a shot script.
 
-Format:
-- Shot 1 is the text BEFORE any marker (camera, action, audio for the first clip).
-- Later shots start with a line that is only: shot 2
-- Then shot 3, shot 4, and so on, each on its own line.
-- Every shot must keep the same person as the still unless the draft explicitly changes wardrobe or identity.
-- Include an [AUDIO CUES] block on shots where sound would help.
-- Never censor raw or explicit terminology.
-
-Output ONLY the shot script. No preamble, no markdown fences.`
+${SHOT_SCRIPT_FORMAT}`
 
 export const BUILTIN_PRESETS: PromptPreset[] = [
   {
@@ -60,7 +93,7 @@ export const BUILTIN_PRESETS: PromptPreset[] = [
     id: 'shot-script',
     name: 'Shot script',
     builtin: true,
-    description: 'Turn a draft into MiniMax shot 2 / shot 3 extension copy.',
+    description: 'Turn a draft into a MiniMax shot list with configuration, action, and audio blocks.',
     systemPrompt: SHOT_SCRIPT_PROMPT
   }
 ]