Ganzflicker: Pseudo-Hallucinatory Visual Phenomena
- Ganzflicker is a full-field visual stimulation method using rapid temporal modulation to induce pseudo-hallucinatory experiences.
- The paradigm minimizes external structure, enabling researchers to probe how low-level visual features evolve into complex, internally generated perceptions.
- Quantitative analyses using topic modeling and sensorimotor norms reveal distinct imagery profiles and neural correlates linked to individual differences.
Searching arXiv for papers on Ganzflicker and closely related flicker phenomena to ground the article. Ganzflicker is a nonpharmacological visual hallucination induction method in which the entire visual field is exposed to a rapidly alternating, spatially uniform flicker. In a large-scale analyzed dataset, the stimulation took the form of a full-screen red/black display flickering at 7.5 Hz for about 10 minutes, under dim lighting, with white noise and instructions to avoid distractions. The paradigm provides highly rhythmic but informationally minimal input: strong temporal stimulation with little structured external content. Under these conditions, participants report pseudo-hallucinatory experiences ranging from simple geometric forms to complex naturalistic scenes. Recent arXiv work treats Ganzflicker as a probe of the generative capacity of the visual system and of individual differences across the imagery spectrum (Chkhaidze et al., 11 Jul 2025).
1. Operational definition and experimental regime
In the cited contemporary literature, Ganzflicker denotes full-field temporal modulation rather than a spatially patterned display. The critical property is spatial uniformity across the visual field combined with rapid alternation over time. The analyzed procedure was straightforward: after exposure to the red/black Ganzflicker stimulus, participants completed demographic items, a single-item visual imagery vividness rating on a 0–10 scale, closed-ended questions, and an open-ended free-text description of what they saw during stimulation (Chkhaidze et al., 11 Jul 2025).
The logic of the paradigm is that the visual system is driven strongly in time while receiving little exogenous scene structure. This suggests that reported percepts are informative about internally generated visual content rather than about ordinary stimulus parsing. In that framework, Ganzflicker functions less as a conventional psychophysical threshold stimulus than as an elicitation method for pseudo-hallucinatory phenomenology.
A recurrent misconception is that Ganzflicker should be understood only in terms of whether flicker is consciously visible. The recent literature instead emphasizes content generation under minimal structured input: the central question is not merely whether the field flickers, but what the visual system produces under that condition.
2. Phenomenology and reported content
Reported Ganzflicker experiences span a broad phenomenological range. At the simpler end are lines, flashes, grids, spirals, balls of light, distortions, and vague patterns. At the more complex end are faces, eyes, cities, forests, hallways, space stations, stars and galaxies, flowers, and immersive scene-like experiences. In the topic-modeling analysis, discovered topics included Lines, Color Flashes, Visual Patterns, Geometric Shapes, Fractals, Spider Web Patterns, Morphing Faces, Forest Scenery, City Skyline, Moving Hallways, Space Travel, Stars & Galaxies, and Rotating Flowers (Chkhaidze et al., 11 Jul 2025).
The dominant empirical contrast is between simple geometric forms and complex naturalistic content. Weak imagers predominantly described simple geometric and low-level visual content, whereas strong imagers reported more semantically organized and naturalistic hallucinations. Importantly, even participants with very weak or absent voluntary imagery often reported seeing something. Ganzflicker therefore does not simply fail in aphantasia-like participants; rather, the content profile shifts.
This phenomenology has been interpreted as evidence for a stratified organization of internally generated visual experience. The simple end of the spectrum plausibly reflects low-level feature-like structure, whereas the naturalistic end reflects more integrated object-, face-, and scene-like organization. The paper’s topic hierarchy was explicitly interpreted as separating a simple geometric/low-level cluster from a structured/naturalistic cluster, with strong imagers drawing more from the latter (Chkhaidze et al., 11 Jul 2025).
3. Quantitative characterization of hallucination content
The main computational analysis treated free-text reports as structured data. From an original sample of 6,664 individuals who underwent the Ganzflicker procedure, the main text analyses retained 4,365 participants after excluding missing or non-English descriptions. For the Lancaster sensorimotor norms analysis, the sample was further reduced to 4,057 participants by excluding descriptions with fewer than three matched normed words. Imagery phenotype was indexed by self-rated vividness from 0 to 10, and for some analyses participants were grouped as weak –$3$, moderate –$7$, and strong –$10$, with final group sizes weak , moderate , and strong (Chkhaidze et al., 11 Jul 2025).
The topic-modeling pipeline used BERTopic on sentence-level documents. Reports were split with NLTK sent_tokenize(), minimally preprocessed, embedded with Sentence-BERT into 384-dimensional sentence vectors, reduced with UMAP using n_components = 10, n_neighbors = 15, and min_dist = 0.1, and clustered with HDBSCAN using minimum cluster size . Topic-defining words were extracted with class-based TF-IDF using the BM25+ variant. The procedure yielded 27 non-outlier topics with $3$0 coherence $3$1, which the authors treated as acceptable for short, noisy, experiential text (Chkhaidze et al., 11 Jul 2025).
At the participant level, each topic was summarized by the maximum topic probability across that participant’s sentences. A Lasso regression using the 27 z-scored topic probabilities explained approximately 5% of the variance in imagery vividness, with $3$2, supplementary $3$3, and optimal $3$4. The strongest positive predictors of vividness were Morphing Faces $3$5, Moving Hallways $3$6, Stars & Galaxies $3$7, City Skylines $3$8, and Forest Scenery $3$9. Negative predictors were Visual Patterns 0, Color Flashes 1, and Lines 2. This pattern quantitatively formalizes the simple-versus-naturalistic contrast (Chkhaidze et al., 11 Jul 2025).
Categorical classification sharpened the same result. One-vs-rest Lasso-regularized logistic classifiers trained on topic probabilities distinguished weak imagery 3 and strong imagery 4, but not the moderate group 5. Weak imagery was defined largely by the absence of naturalistic content; strong imagery showed a much richer topic profile, with 21 of 27 topics surviving regularization and bootstrapping, compared with only 10 of 27 for weak imagery (Chkhaidze et al., 11 Jul 2025).
A separate representational analysis tested whether report differences were better captured by text-only LLMs or vision-LLMs. Descriptions were grouped by the 11 vividness scores, embeddings were averaged within each bin, and an 6 representational dissimilarity matrix was constructed from pairwise Euclidean distances. The theoretical reference structure was effectively
7
Alignment with this structure was strongest for CLIP 8 and SigLIP 9, followed by BERT $7$0, GPT-2 $7$1, RoBERTa $7$2, and BLIP $7$3. This indicates that Ganzflicker reports contain graded visual-semantic structure that vision-LLMs capture particularly well (Chkhaidze et al., 11 Jul 2025).
4. Imagery spectrum and sensorimotor structure
The imagery-spectrum interpretation extends beyond vividness scores. Using Lancaster Sensorimotor Norms, the analysis assigned matched words ratings on 11 sensorimotor dimensions—visual, auditory, gustatory, olfactory, haptic, interoceptive, foot, hand, head, mouth, and torso—plus perceptual strength and action strength composites. Participant-level averages showed visual strength $7$4, haptic $7$5, auditory $7$6, head strength $7$7, hand strength $7$8, perceptual strength composite $7$9, and motor strength composite 0 (Chkhaidze et al., 11 Jul 2025).
The composite regression model was specified as
1
Controlling for description length, both global richness measures predicted vividness: perceptual strength 2; action strength 3; description length 4. At the subdimension level, vividness was associated with visual 5, haptic 6, olfactory 7, and auditory 8; interoceptive 9 was not significant, and gustatory $10$0 was explicitly treated with caution because gustatory terms were rare and likely driven by sparse outliers (Chkhaidze et al., 11 Jul 2025).
Motor-language effects were selective rather than global. Hand $10$1 and head $10$2 predicted vividness, whereas foot, mouth, and torso were non-significant. Mediation analyses using 5,000 nonparametric bootstrap simulations further showed direct effects such as visual $10$3, hand $10$4, and head $10$5 (Chkhaidze et al., 11 Jul 2025).
These findings support a content-based interpretation of the imagery spectrum. Strong imagers’ Ganzflicker descriptions are not merely longer or more visually vivid; they are more sensorimotorically rich and more semantically integrated. A plausible implication is that individual differences in Ganzflicker phenomenology index differences in how internally generated feature-level activity is elaborated into object- and scene-level content.
5. Mechanistic interpretation and relation to flicker theory
The principal interpretation is a layered model of visual generation. On this view, low-level visual areas can generate basic features such as lines, edges, spirals, lattices, and color patches, which could explain why weak imagers and aphantasic individuals still experience Ganzflicker-generated forms. The transition from these fragments to coherent objects, faces, places, and scenes is proposed to require more effective top-down coordination with higher-order regions involved in semantic processing, object recognition, and scene construction, including the fusiform gyrus and broader frontoparietal control networks (Chkhaidze et al., 11 Jul 2025).
This interpretation is explicitly cautious. Ganzflicker hallucinations are not equated with voluntary imagery itself, and the paper does not claim identical mechanisms. It also states that Ganzflicker and Ganzfeld can differ in frequency, complexity, and content, so findings should not automatically be generalized to all induced hallucinations. Additional limitations are substantial: the critical measures are subjective free-text reports, the imagery index is a single self-report vividness item, topic content explains only a modest portion of vividness variance, and some sensorimotor effects—especially olfactory and particularly gustatory—require caution (Chkhaidze et al., 11 Jul 2025).
A related line of flicker research provides a useful physiological constraint. Work on the Talbot-Plateau law shows that flicker-fused stimuli can remain perceptually effective even when time-averaged luminance is matched to a steady background. In particular, under low frequency and ultrabrief flashes, a flicker-fused letter remained recognizable even though it had the same physical luminance as background, and the authors proposed that this anomalous result may reflect unexpected differential activation of ON and OFF retinal channels (Greene et al., 2023). Although that study did not use immersive Ganzfeld or Ganzflicker in the strict sense, it shows that temporal modulation can remain functionally active even when ordinary flicker visibility has disappeared.
This suggests that Ganzflicker should not be reduced to average luminance or to a simple visible-versus-fused dichotomy. Temporal waveform, phase structure, and the balance of bright and dark components may matter for the internal organization of the resulting experience. That inference is compatible with, but not directly demonstrated by, the large-sample Ganzflicker content study.
6. Terminological scope and adjacent usages of “flicker”
Outside perceptual neuroscience, the term flicker is used in several technically distinct ways. In dynamic scene reconstruction with Gaussian splatting, “clip-level flicker” denotes temporal instability, jitter, and visible discontinuities over time caused by inconsistent scene representation across clips (Liang et al., 15 Apr 2026). In display capture and computational photography, “flicker-banding” denotes temporal aliasing between a rolling-shutter sensor and a display’s brightness modulation, yielding color shifts, compound banding, and jagged patterns (Zhou et al., 29 Jun 2026). In video face replacement, flicker refers to temporal instability of the swapped face caused by misregistration and unstable blending boundaries (Wang et al., 2018). In adversarial machine learning, a flickering attack is a spatially uniform but temporally varying global RGB perturbation optimized to fool video classifiers, including over-the-air implementation with a smart RGB bulb (Pony et al., 2020).
These usages are conceptually adjacent because all involve temporally structured modulation, but they are not interchangeable. Ganzflicker is specifically a full-field visual stimulation paradigm used to induce pseudo-hallucinatory experience. Engineering uses of flicker typically refer either to temporal instability artifacts in reconstructed or composited video, or to physically induced brightness modulation captured by imaging systems. Confusing these meanings obscures the distinctive status of Ganzflicker as a probe of internally generated visual content rather than as a rendering artifact, an aliasing artifact, or an adversarial perturbation.
The current literature therefore supports a narrow technical definition: Ganzflicker is a full-field, rapidly alternating flicker paradigm used to study how the visual system generates content under informationally impoverished but temporally strong stimulation. Its main contemporary significance lies in showing that individual differences in imagery are expressed not only in vividness ratings but in the structure, semantic complexity, and sensorimotor richness of induced visual experience (Chkhaidze et al., 11 Jul 2025).