reaDream reports identifying dream video scenes from EEG, with 67% accuracy on a held-out set from one person
reaDream reports that it can use EEG data from one person to select the corresponding dream video scene with 67% accuracy
On July 23, 2026, reaDream founder Ninon Lise Masclef described a test in which EEG recordings taken before awakening were matched against videos from a held-out set. According to Masclef, the system selected the corresponding scene in 67% of cases using data from one adult, which was about 50 times the rate expected from random selection.
EEG records electrical activity in the brain through sensors placed on the scalp. In Ninon Lise Masclef’s presentation at Foresight Vision Weekend, the system receives a recording made before awakening and selects one video from a set prepared in advance. The reaDream website calls this experiment Dream Replay and presents the EEG recording alongside the participant’s account of the dream and the corresponding video scene.
The task is to choose among a known set of options. The algorithm matches the signal against the videos. Performance is measured as the proportion of correct selections and compared with the rate expected by chance. In this context, the phrase “dream decoding” refers specifically to this test: the recording made before awakening must contain a sufficiently distinct signal for the algorithm to associate it with one scene from the set.
In 2013, Tom Horikawa and colleagues published a study in Science that matched brain activity during sleep with responses to visual stimuli during wakefulness and extracted categories of visual imagery. reaDream uses EEG for a similar task, recording signals with sensors placed on the head rather than with an fMRI scanner.
This type of test allows recordings and algorithms to be compared using the same set of scenes. According to reaDream, the EEG signal recorded before awakening was sufficient to match the participant’s dream with the corresponding video more often than random selection would.