Eigen RadarAI
Archive
Analysis

TOPPAN and Keio pursue sharper images from a quantum-and-AI workflow

TOPPAN and Keio University are starting a new phase of research combining quantum computing with generative AI for images. An earlier trial used quantum hardware to produce small character-image cores, then AI to add colour and resolution. The next work targets sharper images and natural-language controls. Advantages in accuracy and generation time on real quantum devices remain research questions.

Artificial Intelligence··Morning
Experimental wired quantum hardware with a display showing a small grayscale character image behind it.

Quantum cores, AI finishing

Japanese technology group TOPPAN Holdings and Japan’s Keio University tested a joint quantum-computing and generative-AI workflow for manga and animation between February 1 and July 31. Keio used flow matching in a quantum circuit to generate a character’s basic structure. TOPPAN then supplied backgrounds, colour and higher resolution using AI.[1], [2]

On a real 10-qubit quantum device, the team generated 16×16-pixel grayscale cores with hairstyle and expression labels across 16 conditions. AI processing turned these into 256×256-pixel colour outputs. The partners thus divide image generation between quantum processing of a small core and conventional AI finishing.[1]

The next phase targets resolution and text controls

TOPPAN announced the next research phase on October 16, with work starting in October. Its targets include higher resolution and specifying image conditions through natural language. The partners also identify three- and four-dimensional scene generation as a longer-term objective. The announcement gives no date for a public production service.[1]

Simulation results and hardware questions

In separate simulations, TOPPAN reports that a smaller circuit reduced operations to about one sixth and increased digit identification accuracy from 16 per cent to 78 per cent. A comparison using 10,000 training examples reported 78.3 per cent accuracy for the quantum model and 54.3 per cent for a similarly sized classical model. Accuracy and generation-time advantages on actual quantum devices remain to be verified.[1]

References

  1. News sourceTOPPAN HoldingsTOPPAN and Keio announce next phase of quantum-assisted image generation↩1↩2↩3↩4
  2. News sourceImpress AI WatchTOPPAN and Keio report quantum-assisted image experiments↩
AI safety and regulationChip wars and the supply chain

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →