FervorCreative AI
Live Latest 21.11.26 · morning 422 tools tracked 1040 workflows indexed 651 topics Hot: Claude Code, Gaussian splatting, ComfyUI

Many polished app animations are just a few still pictures moved around, and image-to-lottie makes that cheap with AI stills, but the craft it encodes is in the checks, one sheet for matching parts, a measured fit and pinned frames, not in the generation.

image-to-lottieLottiedotLottieClaude Codeimage-gencreative-workflowsai-editingprompt-craft

Turn AI Images Into Lottie Animations: A Free Workflow for App Rewards, Gift Boxes and Celebrations

A free tool called image-to-lottie turns a few AI-generated stills into the animation files apps already play. The generation is the easy part. The workflow that makes it look finished is what is worth stealing.

That glossy 3D gift box that wobbles, pops its lid and throws out three coins when you finish a level? There is a good chance nobody animated it in 3D.

A lot of polished app animation is built the cheap way. Someone renders a few still images (the box, the lid, the reward) and then moves those flat pictures around with simple motion: shift, scale, rotate, fade. Done well, your eye fills in the depth. It reads as 3D because the pictures are 3D-looking, not because anything is actually turning in space.

AI image models should make this trivial. They can paint a frosted-glass gift box in seconds. And yet anyone who has tried knows the result usually looks wrong. The lid is a slightly different purple from the box. It sits a few pixels off. The reward pops out in front of the box instead of from inside it. The confetti flies off the edge of the screen.

A small open-source tool published on GitHub on November 19, image-to-lottie, is built around exactly those failures. My position: the tool is useful, but the bigger gift is its list of mistakes and the checks that prevent them. That routine works whether or not you ever install it.

What you get at the end

The output is a Lottie file. Lottie is the lightweight animation format that iOS apps, Android apps and websites already know how to play, which is why product teams use it for loading spinners, success screens and reward moments. A .lottie file (the "dotLottie" version) is a single file with the images packed inside, which is the easiest thing to hand a developer.

image-to-lottie is a free command-line tool (MIT-licensed, so fine for client work) and an add-on skill for AI coding assistants such as Claude Code and Codex. It ships six ready-made motions:

  • gift-open: the box shakes twice, the lid pops off, items rise and fan out.
  • hop: crouch, jump with a wiggle, squash on landing, breathe. Loops.
  • float: a gentle bob and sway with a breathing shadow and glow. Loops.
  • pop-in: springs in with a glow burst, particles and sparkles.
  • fanfare: party poppers in both corners pull back, pop and throw confetti.
  • sequence: a plain flipbook of your own frames, for motion a flat picture cannot fake, like a coin turning.

You need Node.js (version 20 or later) and any image generator that can make transparent backgrounds.

Why the parts never match, and the fix

The author lists the mistakes made while building the example animations in a short file called pitfalls.md. The first one under images is the one that ruins most attempts: parts generated one at a time do not match. Generate the box in one request and the lid in another, and you get two slightly different lights, two cameras, two ideas of scale.

The fix is simple and slightly counterintuitive. Ask for every part on one image, in one request, spaced apart on a transparent background. The model keeps style, lighting, camera and scale consistent within a single picture far better than across separate ones. Then the tool cuts the sheet apart.

The skill ships a prompt template for this. Its four-line requirements block (reflowed here) is worth copying into any multi-part request you make:

One image, about 1536x1024, fully transparent background. The objects do not touch; leave wide transparent gaps between them. Identical camera for all objects: front view, slightly from above. Same lighting and scale. No text, letters, numbers, logos, or ground shadows.

The gift-box prompt adds two details that matter more than they look: the lid must be "exactly as wide as the base's top opening," and the base must be "EMPTY inside," so the rewards have somewhere believable to come from.

The fit is measured, not guessed

The second failure is the lid position. Put it a little too high and a strip of light shows between lid and box, so it doesn't read as closed. Too low and it looks squashed on.

The fit command draws the lid on the base at several overlaps, labelled, on one image. You pick the smallest overlap where the opening is fully hidden. In the project's own example, the candidates ran 130, 156, 182, 200, 217 and 243. At 156 a strip of light still showed; 182 was the first that looked closed, so that went in the settings file.

This is a tiny step, and it is the kind of step that separates "AI made this" from "a designer made this." A designer would never eyeball it twice. Neither should you.

Look at pinned frames, never a screen recording

The third lesson is about checking your work. The author tried judging animations by taking timed screenshots of a live player, and a 0.3-second burst was already gone by the time the screenshot landed, so the animation looked broken when it wasn't.

The tool's answer is two commands. check confirms the file opens, every image is found, the middle and last frames aren't blank, and warns (with frame numbers) when something touches the canvas edge or when an image is far bigger than it will ever be shown. render draws exact frames you name onto one contact sheet, so you can see the hero moment, the frames around it, and the last frame side by side.

The skill gives a short pass list: nothing important cropped by the edge, a clean final frame (on one-shot animations that frame is what users keep staring at), first and last frames matching on loops, and items emerging from inside the box rather than in front of it.

Put this into practice: your first gift box in an afternoon

Here is the routine, end to end. You can let an AI coding assistant run the commands for you with the skill installed, or type them yourself.

1. Settle the look first. Find one to three reference images or name a specific style before you generate anything. The pitfalls file admits several full redesigns happened because the look was guessed.

2. Generate one part sheet. Use the template: base on the left, lid in the middle, one reward on the right, transparent background, same camera. Look at it. If the lid is visibly the wrong width or the base isn't empty, regenerate; the author's rule is "regenerate rather than fix a bad sheet."

3. Install the tool. The README's recommended route is installing the packaged download file from the project's Releases page with one npm command (npm is the installer that comes with Node.js). The pitfalls file warns that npm's github: shortcut breaks, so use that release file, or the npx route, which runs the tool without installing it.

4. Size the art, then cut the sheet. Keep images at about twice the size they will be shown on screen. The pitfalls file records a 768-pixel image shown at 200 pixels producing a 7 MB file.

image-to-lottie split sheet.png -o parts --names base,lid,item

If it complains about the background or the number of objects, the sheet needs regenerating.

5. Find the fit.

image-to-lottie fit parts/base.png parts/lid.png -o fit.png

Open fit.png and pick the smallest overlap that hides the opening.

6. Build. Put that overlap in a small settings file alongside the gift-open motion and your three parts (the README's example file shows the exact format), then:

image-to-lottie build config.json -o out/anim.json --lottie out/anim.lottie

7. Check and render a contact sheet.

image-to-lottie check out/anim.lottie
image-to-lottie render out/anim.json --frames first,mid,last --sheet out/sheet.png

Look at the sheet. For the gift box, the skill tells you to also look at the frames around the pop (frames 50 to 76 at 60 frames per second). If check warns that an image is far too large, resize it and rebuild.

8. Ship the single file. The .lottie file is the easiest hand-off to a developer, because the images travel inside it.

Small fixes don't need a rebuild. inspect lists the layers and edit changes one layer's position, scale, rotation, opacity or timing, so "make the glow softer" is one command.

Where it breaks

It is one day old with one star. A single developer published it on November 19. The examples are real, but you are an early tester.

Flat pictures have a ceiling. Anything that needs to genuinely turn, morph or flip in space can't be faked by moving a still. The tool's answer is a flipbook of frames (its spin command makes coin-style turning frames from one front image), and frames make files heavier.

The motions are presets. Six of them. If your brand needs a motion that isn't a box opening, a hop, a float, a pop-in or a celebration, you're writing a new one or animating it yourself. The README credits LottieFiles' motion-design skill and diffusionstudio's text-to-lottie for shaping its motion rules and says they pair well: they cover vector animation, this covers flat picture parts.

The look is only as consistent as your sheet. One sheet helps a lot. It does not guarantee your second animation matches your first. Keep the same reference images and prompt wording across a set.

Agents can overstep. A quirky but real warning from the pitfalls file: if you let one AI assistant generate images inside a project folder with "commit after every change" rules, it may try to save unreviewed art into version control. The author's fix is a plain instruction: "Only write the named file. Do not run git or edit anything else."

Image rights still apply. Whatever generator you use for the parts, its terms decide whether you can ship the result in a commercial app. The tool doesn't change that.

The routine is the product

The interesting thing about image-to-lottie is how little of it is about AI. The generation step is one prompt. Everything else is the patient, checkable craft a motion designer would apply anyway: make the parts together, measure the fit, look at the exact frame where the action happens, keep files the right size.

AI made the stills almost free. That moved the value to the checks. Copy the requirements block into your next multi-part image prompt, even if you never animate it, and see how much better the pieces fit together.

Then make one gift box. If the lid closes cleanly on your contact sheet, you have a reward animation your developer can drop in this week.


From the briefing