Flows in Mediafy Studio: Chain Image, Video and Voice

Here's the difference a good description makes. Say you type this into Magic, the tool that builds a flow from plain language:

Make a video about my coffee shop.

That's too thin. Nothing says what the image should show, how it should move, or what the voice should say. Compare:

Image: a flat white on a wooden counter, morning window light, steam rising, shallow depth of field, vertical. Video: slow push-in on the cup, steam drifting. Voice-over: a warm, calm voice saying "Slow mornings start here. Fresh roasted, poured just for you."

The second version names three steps, what each one produces, and how they connect. That's what an AI workflow canvas needs from you.

What a flow is

A flow in Mediafy Studio is a node canvas. Each node is one step, such as an image, a video or a voice-over. You connect them so the output of one feeds the next, then run the whole chain in one go.

Doing this by hand works fine. You make an image in the Studio, download it, upload it to a video model, then make audio separately. A flow just removes the shuffling between steps. It pays off most when you repeat the same pattern, like a weekly product teaser or a set of social clips.

When to use a flow, and when not to

Use a flow when:

• You already know the sequence (image, then video, then voice).
• You'll make several variations of the same pattern.
• You want one place to see how the pieces connect.

Skip it when you're still exploring. If you don't know what look you want, use the regular Studio, try a few prompts and models, and come back to a flow once you've found something that works. Flows are better for repeating a recipe than for finding one.

Build your first flow in five steps

1. Sketch the chain on paper first. One line per step: "still image of the cup", "animate it 5 seconds", "voice-over line". If you can't write the line, you aren't ready to build the node.
2. Describe it to Magic. Write the chain in plain language, as in the example above. Name the subject, the motion and the spoken words. Magic builds a starting canvas for you.
3. Check every node. Magic gives you a draft, not a finished product. Open each node and read the prompt, model and settings. Fix anything vague.
4. Run it small. Test with cheaper choices before you commit to the best model.
5. Save what works. Keep the winning prompts in My Prompts, and find your results in Assets.

Write each node for its own job

Each step has a different job, so each prompt should be written differently.

The image node: build a good first frame

The video can only animate what the image gives it. Use the subject, setting, light and lens structure from AI Image Prompt Structure: Subject, Setting, Light, Lens. Leave some room in the frame for movement. Don't pack the image with ten details if the video model will only move one of them.

If the image needs text on it, check How to Get Readable Text Inside AI Images first. A garbled sign in the first frame carries into every frame after it.

The video node: describe motion, not the scene

The image already shows the scene. The video prompt should say what changes. Compare:

• Weak: A cup of coffee on a counter.
• Better: Slow push-in toward the cup. Steam curls upward and drifts left. Background stays soft and still.

That's the whole idea behind AI Video Prompts: Describe the Motion, Not Just the Scene. For specific moves, see Camera Movement Prompts for AI Video: Dolly, Pan, Orbit.

The voice node: write for the ear

Spoken lines read differently from written ones. A few habits that help:

• Keep it short. Match the words to the clip length, so a 5-second clip gets one short sentence, not a paragraph.
• Read it aloud before you run it. If you stumble, the voice will too.
• Spell out awkward things. Write "twenty percent off" instead of "20% off" if you want it read a certain way.
• Add delivery notes in plain words, like "warm, unhurried", where the node accepts a description of the voice.

Keep the same person across steps

If your flow features a person, a spokesperson or a recurring character, don't describe their face from scratch in each prompt. Design a character once and reuse it. In the Studio you can type @ to bring in a saved character or a Seed. Seeds are saved sets of reference images that hold a look steady. They guide the look but aren't LoRA training, and the studio doesn't train custom models. The full approach is in Keep a Consistent AI Character Across Images and Video.

Mistakes people actually make

• Cramming everything into one prompt. A long prompt that asks the image node to also handle motion and speech muddles all three. Give each node only its own job.
• Wrong shape for the platform. If the image is square and you wanted a vertical clip, the video will crop or letterbox. Decide the ratio before the first node. Aspect Ratio for Social Media covers which one fits where.
• Running the expensive version first. Errors in step one flow through every later step. Prove the chain with a cheap draft before you pay for the best model at each node.
• Not reading the Magic draft. It's a good starting point. It isn't a guarantee that every prompt says what you meant.
• Copying negative prompts everywhere. Not every model reads them. Check Negative Prompts: When They Help and Who Ignores Them before you rely on them.
• A voice line longer than the clip. The speech gets cut off or the pacing feels rushed. Trim the words, not the video.

Keep the cost under control

Credits are pay-as-you-go, and each model has its own cost. In the Studio, the cost shows on the Generate button before you run. A chain has several steps, so add them up. For a rough sense of scale, a Nano Banana image is 3 credits. Video models vary a lot, so check before you run, not after.

A practical routine: test the chain with a fast, cheap image model first. When the framing and motion idea work, switch the image node to a stronger model and run it once for real. Failed generations are refunded automatically, but a result that runs fine and just isn't what you wanted is not a failure, so drafts are worth the small cost. The details are in How Mediafy Credits Work: Costs, Packs and Refunds.

A second example: a service business

A plumber, a salon, a bakery: the pattern is the same. Try this Magic description:

Image: a tidy van parked outside a bright suburban house, early morning, friendly and clean, vertical. Video: slow dolly in toward the front door, a gentle breeze moving the trees. Voice-over: "Same-day repairs, fair prices. Call before noon and we'll be there today."

Run it, check the result, then swap the voice line and the image for the next campaign. The structure stays; only the details change. That's the real value of the canvas.

Where to go from here

If you haven't made an image in the Studio yet, start with Mediafy Studio Tutorial: Your First Image in 5 Minutes. If you want a structured path through the whole process, the AI Creator Lab course covers it.

When you're ready to build your first chain, sign in with Google and try a flow.