--- title: Stills to motion | FLORA API description: Branch from one Technique into another, carrying selected outputs as inputs. --- Once you’ve nailed the stills, branch into a different Technique using the winners as keyframes — without losing context. This is where FLORA MCP starts paying for itself: fast iteration in chat, then background mode for the longer-running motion work. **Prerequisite:** A set of approved stills from [Discover and run](/mcp/recipes/discover-and-run/index.md) or [Iterate on favorites](/mcp/recipes/iterate-on-favorites/index.md). ## What you type > Take the winners and run them through the Social Motion Technique. I need 3-second loops for IG. ## What the agent does 1. Switches to background mode Motion generation takes longer than stills. Most agents propose moving the task to a background panel — Claude calls this **Cowork**, Cursor calls it an **agent run**, and so on. The task runs while you keep working in chat. 2. list\_techniques + retrieve\_technique The agent finds the Social Motion Technique and reads its inputs. It expects keyframes as `imageUrl` inputs and accepts `duration` and `loop` parameters. 3. create\_technique\_run The agent submits one run per chosen still, passing each as a keyframe. The background panel displays the task: “Generating motion versions of 5 thumbnails.” 4. retrieve\_technique\_run The agent polls each run. Inline MP4 previews start playing on autoplay-loop as each one finishes, labeled with the source thumbnail number. ## What you see A task panel appears with progress per run. As each motion version finishes, it pops into the chat as a short looping MP4, tagged “from #4”, “from #7”, and so on. Click any one to open it full-size or download. ## Why context survives the branch You didn’t re-describe the brand direction, the brief, or your favorites. The Social Motion Technique didn’t need to know any of that — the agent just passed the chosen output URLs as keyframes. The Technique system handles the rest. This is the part that’s hard with single-shot generative MCPs. With FLORA, **your choice of favorites doesn’t get lost when you switch from stills to motion.** ## Variations - *“Run only #4 through Social Motion — 3 seconds, loop. And #7 through the Hero Motion Technique — 6 seconds, no loop.”* - *“Use #4 as the start frame and #7 as the end frame for a single morphing loop.”* (Requires a Technique that accepts paired keyframes.) - *“Generate the motion versions in 9:16 vertical for IG Stories.”* (Most motion Techniques expose an aspect ratio input.) ## Under the hood ``` # The agent finds the motion Technique list_techniques(query="social motion") retrieve_technique(slug="social-motion") # One run per chosen still, in parallel create_technique_run( slug="social-motion", inputs={ "keyframe": "", "duration_seconds": 3, "loop": true } ) create_technique_run( slug="social-motion", inputs={ "keyframe": "", ... } ) # ...etc for each chosen still # Each run polled independently retrieve_technique_run(runId) ``` ## When to use background mode vs chat | | Inline chat | Background mode | | ------------------------ | --------------------------------- | ---------------------------------------------------- | | Run time | Seconds to \~30s | Anything longer | | Inline rendering | Yes | Yes, when ready | | Keep working in parallel | Blocked while polling | Unblocked | | Best for | Discovery, iteration, single runs | Motion, batches, anything >30s | | Examples by client | Default chat | Claude Cowork, Cursor agent runs, VS Code agent mode | Most clients auto-pick background mode for longer tasks. You can override: *“Do this inline — I want to watch it.”* ## Tips - For consistent motion across a set, run all keyframes through the same Technique in one batch — don’t mix Techniques mid-set. - Motion Techniques charge more credits per run than stills. Confirm cost in `retrieve_technique` before kicking off a big batch. - Save approved motion outputs to a Project (`attach_project_asset`) so they show up in FLORA’s canvas alongside the stills.