A damaged print restored and colourised by Flux Kontext, then set in motion for five seconds by Kling V3: a blink, a smile, a slight turn of the head. The full workflow, no account needed.
Real workflow, real results. Pan and zoom freely.
Use this workflowThe real workflow: the old photo enters an Image node that restores it, and the restored portrait becomes the first frame of a Video node that brings it to life.
At a glance
A grandmother blinking, a great-grandfather smiling: watching a seventy-year-old photo move is the thing that makes everyone cry, and it takes two steps. The first restores and colourises the print with Flux Kontext, an editing model that repairs what it is given instead of inventing. The second animates the restored image with Kling V3, five seconds from a first frame.
Each step has its lock, and the demo shows them in full. The restoration names what must not change, features, expression, hairstyle, clothes, framing: without that list you get a sharper stranger. The animation forbids everything but the face, static camera, no zoom or cut, hands still: without that list the portrait becomes a film shot and the memory is lost. The movements stay small, which is exactly what moves people.
The demo photo is generated, so as to animate nobody's face. Yours enters through the Media node. The full run costs about a dollar and a quarter, six credits of which go to the restoration: the video is most of the price, restore first and only animate once the face is right.
Import the print as it is
Photograph it or scan it, folds and stains included: the restoration takes care of them. A sharp, front-facing face is best, the model only repairs what it sees.
Adapt the prompt's colours
The restoration prompt describes the demo's dress, hair and light. Write yours if you know them, otherwise let the model pick natural colours. The lock on the features does not change.
Check the likeness before animating
Compare the restored portrait to the original, eyes and jaw first. If it is not the right person, fix the prompt and re-run the restoration alone: it costs six credits, the video a hundred.
Run the animation
The Video node starts from the restored portrait. Keep the movements small, a blink, a smile, a breath. The render downloads or goes to the edit.
The same canvas, tuned for different needs: each variation is two or three lines of prompt away.
The demo case, and the most moving: a posed, front-facing portrait where the person looks at the lens. A blink and a smile are enough. Send the video to the family with the original photo next to it, the contrast is what does the work.
Two people side by side: the animation prompt names both, "he turns his head toward her, she smiles". The more faces there are, the smaller the movements must stay, or Kling invents some.
You know the colours: write them into the restoration prompt, the colour of the jumper, the wall, the eyes. The colourisation becomes a restitution rather than an invention.
If the movement does not interest you, the restore an old photo workflow stops at the image, with a before-and-after restoration video rather than an animation of the face.
Several animated photos, one per decade, chained in the Montage with music. Each animation is a node, the montage assembles them and exports a single file.
Too many faces for a faithful animation: restore only, or animate a crop on one or two people. The Media node accepts an already cropped photo.
The video costs fifteen times the restoration. If the restored portrait is not the right person, the video will animate the wrong one. Look at the image, then only run the Video node.
"She stands up and walks toward the camera" turns the portrait into an invented scene: Kling has to create a body, a room, a movement, and nothing is faithful any more. A blink, a smile, a breath.
Without "static camera, no zoom, no cut", the model adds a dolly move or a reframing, and the print becomes a film shot. The photo must stay a photo that breathes.
The model picks plausible colours, not the real ones. If you know the dress was red, say so: a wrong colour on a memory shows more than a fold.
Yes, by cabling the Media node straight into the Video node, but the result inherits the scratches and the sepia, and Kling animates those too. Restoration costs six credits out of a hundred: it is always worth it.
That is the job of the prompt's lock, and the reason for the checking step. On the demo, the restoration keeps the features but colourises freely; the animation only touches the face. Always compare before animating.
Not in this workflow, which keeps the mouth closed to stay faithful. Making a face talk needs a voice and another model: that is the talking avatar workflow, which starts from an image and a text.
One hundred and seven credits, six for the restoration and about a hundred for five seconds of Kling V3 in 1080p. The price shows on each node before you click. Re-running the restoration alone only costs six credits.
A scan or a photo of the print where the face is sharp. The print's defects, folds, stains, yellowing, are exactly what the restoration fixes. A blurry face, however, stays blurry and is then animated blurry.
Five seconds in the demo, in the photo's format. Enough for a blink, a smile and a breath, and it is what shares best. The node accepts longer, at a proportional price.
Yes, and it is often the simplest: photograph the print with the phone, import it into the Media node from the mobile canvas. The creating on mobile guide explains the gestures.
No. What you upload stays in your workspace and is handed to no training. The question matters more here than elsewhere, since it is your relatives.
Bring an old photo to life with AI
Use this workflowFree account, no card required. The workflow opens pre-filled in your canvas.
Flux Kontext
Kling V3
Topaz Precision