{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "kang-et-al-gigagan-scaled-text-image-synthesis-2023", "title": "GigaGAN: scaling a fast generator without losing editable image styles", "summary": "Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods.", "boundary": "Original timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.", "area": "Adversarial image generation", "topic": "Artificial intelligence", "year": 2023, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "kang-et-al-gigagan-scaled-text-image-synthesis-2023", "target": { "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "content_hash": "sha256:ec43fa7121db68a9e1b5432e901b2b1dae11a4f246a6334acb373eb1e0c1c1ab" }, "humans": [ "Minguk Kang", "Jun-Yan Zhu", "Richard Zhang", "Jaesik Park", "Eli Shechtman", "Sylvain Paris", "Taesung Park" ], "authors_complete": true, "human_role_note": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261017-arxiv-2303-05511v1", "title": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1", "date": "2023-03-09" } ], "points": "100000", "published_grade": { "id": "kang-et-al-gigagan-scaled-text-image-synthesis-2023", "entry_id": "kang-et-al-gigagan-scaled-text-image-synthesis-2023", "band": "A", "modifier": null, "target": { "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "content_hash": "sha256:ec43fa7121db68a9e1b5432e901b2b1dae11a4f246a6334acb373eb1e0c1c1ab" }, "what_changed": "Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods.", "scope": "Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods. Original timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "content_hash": "sha256:ec43fa7121db68a9e1b5432e901b2b1dae11a4f246a6334acb373eb1e0c1c1ab" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-23T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Minguk Kang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jun-Yan Zhu", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Richard Zhang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jaesik Park", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Eli Shechtman", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Sylvain Paris", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Taesung Park", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Original timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.", "domain": "Adversarial image generation" }, "sources": [ { "citation": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1" } ], "statement": "# GigaGAN: scaling a fast generator without losing editable image styles\n\n## Why make a bigger generator?\n\nImagine a drawing machine with one set of brushes for every possible prompt. Making every brush wider is expensive and can make training unstable. GigaGAN instead learns a bank of filters and mixes them according to the current example. A filter is a small numerical pattern used to transform image features. The mixture supplies different effective filters for different prompts without repeating every candidate filter operation at every image location.\n\n## Small pictures get several kinds of feedback\n\nThe generator first creates a small image. Attention lets distant image regions exchange information and lets image features consult words in the prompt. The discriminator, which learns to distinguish generated and training images, examines multiple image scales. Extra losses encourage image–text agreement. These ingredients work together; the paper’s ablations do not show that every addition improves every quality metric.\n\n## A sharper picture can contain invented detail\n\nA separate upsampler enlarges the base image while consulting text. In a classroom example, enlarging a square from 64 pixels per side to 512 multiplies the side length by eight and the pixel count by sixty-four. It does not create sixty-four times as much observed information. A tiny blurred object may support several plausible textures. The upsampler chooses detail from learned patterns rather than recovering a uniquely measured hidden original.\n\n## Fast is a condition, not a complete verdict\n\nThe source reports particular A100 generation times. Training and the choice of upsampling stages are separate costs. Its distributional image score can beat a comparator while its pictures remain less convincing: the authors explicitly show realism and composition failures. Think of the score as one test of a collection of images, not a certificate that each caption was obeyed.\n\n## Leave a comparison somebody can audit\n\nAsk your agent to compare one base-generation row and one upsampling row in the original paper. Keep input size, output size, hardware, stage, quality metric and missing conditions in separate columns. Mark your own 64-to-512 arithmetic as a teaching calculation. An informative partial result can explain why two apparently impressive timing numbers cannot yet be compared fairly.\n\n## Scientific scope\n\nOriginal timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.\n\n## Sources for this explanation\n\n- [Scaling up GANs for Text-to-Image Synthesis](https://arxiv.org/abs/2303.05511v1): Original main sections3–5, Tables1–5 and Figure9; appendix experiments and independent runtime reproduction not claimed.\n", "summary": "Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods.", "title": "GigaGAN: scaling a fast generator without losing editable image styles" }, "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "source_ref": { "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "content_hash": "sha256:ec43fa7121db68a9e1b5432e901b2b1dae11a4f246a6334acb373eb1e0c1c1ab" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:2cf491db9c0be3d06f1f3df35863d0df17566b76c5add7f3245af6ee07baebc2" }, "target": { "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "content_hash": "sha256:ec43fa7121db68a9e1b5432e901b2b1dae11a4f246a6334acb373eb1e0c1c1ab" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "kang-et-al-gigagan-scaled-text-image-synthesis-2023", "entry_id": "kang-et-al-gigagan-scaled-text-image-synthesis-2023", "revision": 1, "title": "GigaGAN: scaling a fast generator without losing editable image styles", "context": "Adversarial image generation", "track": "Reusable scientific method, experimental result or historical framework", "band": "A", "modifier": null, "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-23", "evidence_cutoff": "2026-10-23", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods.", "scope": "Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods. Original timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.", "evidence_summary": "The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. Original timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.", "references": [ { "title": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1", "locator": "Original scoped contribution, statement/method and evidence limits; source-specific attribution retained." } ], "roles": [ { "who": "The named authors in the research record", "credit": "SCIENCE", "did": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred." } ], "target": { "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "content_hash": "sha256:ec43fa7121db68a9e1b5432e901b2b1dae11a4f246a6334acb373eb1e0c1c1ab" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "100000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": { "id": "kang-et-al-gigagan-scaled-text-image-synthesis-2023", "target": { "record_id": "urn:uuid:4c97347a-e7a0-4dda-aa51-a1cbdcf46891", "version": 1, "content_hash": "sha256:ec43fa7121db68a9e1b5432e901b2b1dae11a4f246a6334acb373eb1e0c1c1ab" }, "sections": [ { "title": "Why make a bigger generator?", "text": "Imagine a drawing machine with one set of brushes for every possible prompt. Making every brush wider is expensive and can make training unstable. GigaGAN instead learns a bank of filters and mixes them according to the current example. A filter is a small numerical pattern used to transform image features. The mixture supplies different effective filters for different prompts without repeating every candidate filter operation at every image location." }, { "title": "Small pictures get several kinds of feedback", "text": "The generator first creates a small image. Attention lets distant image regions exchange information and lets image features consult words in the prompt. The discriminator, which learns to distinguish generated and training images, examines multiple image scales. Extra losses encourage image–text agreement. These ingredients work together; the paper’s ablations do not show that every addition improves every quality metric." }, { "title": "A sharper picture can contain invented detail", "text": "A separate upsampler enlarges the base image while consulting text. In a classroom example, enlarging a square from 64 pixels per side to 512 multiplies the side length by eight and the pixel count by sixty-four. It does not create sixty-four times as much observed information. A tiny blurred object may support several plausible textures. The upsampler chooses detail from learned patterns rather than recovering a uniquely measured hidden original." }, { "title": "Fast is a condition, not a complete verdict", "text": "The source reports particular A100 generation times. Training and the choice of upsampling stages are separate costs. Its distributional image score can beat a comparator while its pictures remain less convincing: the authors explicitly show realism and composition failures. Think of the score as one test of a collection of images, not a certificate that each caption was obeyed." }, { "title": "Leave a comparison somebody can audit", "text": "Ask your agent to compare one base-generation row and one upsampling row in the original paper. Keep input size, output size, hardware, stage, quality metric and missing conditions in separate columns. Mark your own 64-to-512 arithmetic as a teaching calculation. An informative partial result can explain why two apparently impressive timing numbers cannot yet be compared fairly." } ], "references": [ { "title": "Scaling up GANs for Text-to-Image Synthesis", "url": "https://arxiv.org/abs/2303.05511v1", "locator": "Original main sections3–5, Tables1–5 and Figure9; appendix experiments and independent runtime reproduction not claimed." } ] }, "reading_status": "EXPLANATORY_GUIDE" }