Gemini 3.1 Flash Image
Model Card
                         Gemini 3.1 Flash Image - Model Card

Model Cards are intended to provide essential information on Gemini models, including known
limitations, mitigation approaches, and safety performance. Model cards may be updated from
time-to-time; for example, to include updated evaluations as the model is improved or revised. See the
Google DeepMind site for a comprehensive list of model cards.

Published / Model Release: February 2026

                                        Model Information

Description: Gemini 3.1 Flash Image is a member of the Gemini series of models, a suite of
highly-capable, natively multimodal reasoning models. Gemini 3.1 Flash Image can comprehend input
from different information sources, including text, images, audio and video. Image and text output is
generated in the response.

Model dependencies: Gemini 3.1 Flash Image is based on Gemini 3 Flash.

Inputs: Text strings (e.g., a prompt, document(s)) and images, with a token context window of up to 1M.

Outputs: Image, with a 4K token output

Architecture:: Gemini 3.1 Flash Image is based on Gemini 3 Flash. For more information about the model
architecture for Gemini 3 Flash, see the Gemini 3 Flash model card.

                                                                                                          1
                                             Model Data

Training Dataset: Gemini 3.1 Flash Image is based on Gemini 3 Flash. For more information about the
training dataset for Gemini 3 Flash, see the Gemini 3 Flash model card.

Training Data Processing: For more information about the training data processing for Gemini 3.1 Flash
Image, see the Gemini 3 Flash model card.

                              Implementation and Sustainability

Hardware: Gemini 3.1 Flash Image was trained using Google’s Tensor Processing Units (TPUs). TPUs are
specifically designed to handle the massive computations involved in training LLMs and can speed up
training considerably compared to CPUs. TPUs often come with large amounts of high-bandwidth
memory, allowing for the handling of large models and batch sizes during training, which can lead to
better model quality. TPU Pods (large clusters of TPUs) also provide a scalable solution for handling the
growing complexity of large foundation models. Training can be distributed across multiple TPU devices
for faster and more efficient processing.

The efficiencies gained through the use of TPUs are aligned with Google's commitment to operate
sustainably.

Software: Training was done using JAX and ML Pathways.

                                             Distribution

Gemini 3.1 Flash Image is distributed in the following channels; respective documentation shared in line:

    ●​   Gemini App
    ●​   Gemini Enterprise Agent Platform
    ●​   Google AI Studio
    ●​   Gemini API
    ●​   Google AI Mode
    ●​   Google Antigravity
    ●​   Google Slides
    ●​   Google Vids
    ●​   Google Flow
    ●​   NotebookLM
    ●​   Stitch

                                                                                                            2
                                             Evaluation

The following Evaluation approach and results are for Gemini 3.1 Flash Image. For more information
about the evaluation for Gemini 3 Flash, see the Gemini 3 Flash model card.

Approach: Gemini 3.1 Flash Image was evaluated using the methodology below:

    ●​ Human evaluations of several different quality aspects of image generation were conducted.
       Prompt sets for human evaluations were in two broad categories:
          ○​ New capability sets: diverse Text-to-Image (T2I) and editing evals are curated, covering
              a wide range of capabilities, e.g. Text, Style, Character, Reasoning, Factuality,
              Infographics, Ink (doodle) based Editing, Multi-Turn, Multi-Image (Multi-Product
              Recontextualization, Multi-Character, etc.)
          ○​ Regression sets: popular use cases observed on Gemini 2.5 Flash Image and Gemini 3
              Pro Image, to ensure Gemini 3.1 Flash Image does not show noticeable regression.
    ●​ Capabilities / Benchmarks:
          ○​ T2I: General Text-to-Image, Text Rendering, Visual Design.
          ○​ Editing: General Image Editing, Stylization, Character Editing, Object/Environment
              Editing, Factuality, Infographics, Ink (doodle) based Editing, Multi-Image (Multi-Product
              Recontextualization, Multi-Character, etc.), Multi-Turn.
          ○​ Public benchmark: a subset of GenExam.
    ●​ Eval Methodology
          ○​ SxS human eval to get Elo across diverse T2I, Editing, and Multi-Turn.
          ○​ Single sided expert rating on subject-focused tasks on EDU tasks, GenExam.
          ○​ AutoRater on factuality, style diversity of non-natural images.

                                                                                                      3
Results: Results for Gemini 3.1 Flash Image are below.

Capabilities: Text-to-Image

Capability          Gemini 3.1      Gemini 3.1      Gemini 2.5     Gemini 3        GPT-Image      Seedream       Grok
Benchmark           Flash Image     Flash Image     Flash Image​   Pro Image​      1.5            5.0 Lite       Imagine
                                                                   (“Nano                                        Image Pro
                    (Thinking +                     (“Nano
                                                                   Banana Pro”)
                    Text Search +                   Banana”)
                    Image Search)

Overall
Preference          1079.0 ± 7.0    1073.0 ± 5.0    942.0 ± 6.0    1021.0 ± 5.0    1047.0 ± 5.0   928.0 ± 8.0    906.0 ± 6.0
(GenAI-Bench)

Visual Quality
                    1140.0 ± 6.0    1129.0 ± 6.0    929.0 ± 6.0    1043.0 ± 5.0    975.0 ± 5.0    759.0 ± 10.0   953.0 ± 5.0
(GenAI-Bench)

Infographics
                    1114.0 ± 14.0   1074.0 ± 12.0   881.0 ± 13.0   1102.0 ± 13.0   985.0 ± 12.0   890.0 ± 22.0   942.0 ± 21.0
(Factuality)

Capabilities: Editing

Capability          Gemini 3.1      Gemini 3.1      Gemini 2.5     Gemini 3        GPT-Image      Seedream       Grok
Benchmark           Flash Image     Flash Image     Flash Image​   Pro Image​      1.5            5.0 Lite       Imagine
                                                                   (“Nano                                        Image Pro
                    (Thinking +                     (“Nano
                                                                   Banana Pro”)
                    Text Search +                   Banana”)
                    Image Search

General Editing     1065.0 ± 9.0    1047.0 ± 9.0    913.0 ± 9.0    1051.0 ± 10.0   995.0 ± 8.0    937.0 ± 9.0    989.0 ± 8.0

Character
                    1056.0 ± 7.0    1049.0 ± 7.0    952.0 ± 7.0    1050.0 ± 8.0    1025.0 ± 7.0   894.0 ± 8.0    972.0 ± 7.0
Editing

Creative            1023.0 ± 7.0    1031.0 ± 7.0    976.0 ± 7.0    1004.0 ± 7.0    1017.0 ± 7.0   938.0 ± 7.0    1016.0 ± 7.0

Object/Environ
                    1029.0 ± 8.0    1018.0 ± 8.0    945.0 ± 8.0    1042.0 ± 10.0   976.0 ± 8.0    946.0 ± 9.0    1022.0 ± 8.0
ment Editing

Multi-Input (1-3)   1037.0 ± 8.0    1016.0 ± 8.0    919.0 ± 9.0    1056.0 ± 12.0   1014.0 ± 9.0   951.0 ± 9.0    N/A

Stylization         1045.0 ± 7.0    1031.0 ± 7.0    862.0 ± 8.0    1045.0 ± 9.0    996.0 ± 7.0    984.0 ± 7.0    1021.0 ± 8.0

                                                                                                                                4
                                Intended Usage and Limitations

Benefit and Intended Usage: Gemini 3.1 Flash Image is capable of using Gemini’s real-world knowledge
to deliver precise results and reflect the world around you, from complex infographics to historically
accurate scenes. It is well-suited for applications that require:

    ●​ creation and editing of images with professional levels of precision and control and multiple,
       quick iterations
    ●​ generation of clear text for posters and intricate diagrams
    ●​ long context real-world knowledge
    ●​ localized text rendering across several languages
    ●​ studio-quality control

Known Limitations: Gemini 3.1 Flash Image may exhibit some of the general limitations of foundation
models, such as hallucinations. There may also be occasional slowness or timeout issues.

Gemini 3.1 Flash Image still has room for several quality improvements:

    ●​   Text rendering: poor in small text (often blurry in 1k model), long paragraphs, page length
    ●​   Character consistency is not always perfect between input images and generated output image
    ●​   Masked/Doodle based editing: partial instruction following and persistent ink
    ●​   When editing images: infrequent copying/pasting from user's input image to generated image
    ●​   Occasional confusion around spatial localisation (e.g. left/right etc.)
    ●​   Still limited in advanced capabilities with world knowledge, 3D reasoning and factuality

The knowledge cutoff date for Gemini 3.1 Flash Image was January 2025.

Acceptable Usage: For more information about the acceptable usage for Gemini 3.1 Flash Image, see
the Gemini 3 Flash model card.

                                   Ethics and Content Safety

Evaluation Approach: Gemini 3.1 Flash Image was developed in partnership with internal safety, and
responsibility teams. A range of evaluations and red teaming activities were conducted to help improve
the model and inform decision-making. These evaluations and activities align with Google's AI Principles
and responsible AI approach, as well as Google's Generative AI policies (e.g. Gen AI Prohibited Use
Policy and the Gemini API Additional Terms of Service). As Gemini 3.1 Flash Image is based on Gemini 3
Flash, see the Gemini 3 Flash model card for additional Ethics & Content Safety details.

                                                                                                        5
Evaluation types included but were not limited to:

    ●​ Training/Development Evaluations including automated and human evaluations carried out
       continuously throughout and after the model’s training, to monitor its progress and
       performance;
    ●​ Human Red Teaming conducted by specialist teams across the policies and desiderata,
       deliberately trying to spot weaknesses and ensure the model adheres to safety policies and
       desired outcomes;
    ●​ Trust Assurance Evaluations conducted by evaluators who sit outside of the model
       development team, used to independently assess responsibility and safety governance
       decisions;
    ●​ Ethics & Safety Reviews were conducted ahead of the model’s release.

Safety Policies: Gemini’s safety policies are based on Google’s standard framework, which aim to
prevent our Generative AI models from generating harmful content, including:

    1.​ Content related to child sexual abuse material and exploitation
    2.​ Hate speech (e.g. dehumanizing members of protected groups)
    3.​ Dangerous content (e.g. promoting suicide, or instructing in activities that could cause
        real-world harm)
    4.​ Harassment (e.g. encouraging violence against people)
    5.​ Sexually explicit content
    6.​ Medical advice that runs contrary to scientific or medical consensus

We continue to improve our internal evaluations, including refining automated evaluations to reduce
false positives and negatives, as well as update query sets to ensure balance and maintain a high
standard of results.

Frontier Safety Assessment: Gemini 3.1 Flash Image is part of the Gemini 3 family of models. We
evaluated Gemini 3.1 Pro for frontier safety as it was the most generally capable model as of publication
of this model card, and it did not reach any Critical Capability Levels (CCLs) outlined in our Frontier
Safety Framework. Our assessments have shown that Gemini 3.1 Flash Image is likely to perform similarly
to Gemini 3.1 Pro, therefore based on Gemini 3.1 Pro, we are confident that that Gemini 3.1 Flash
Image is also unlikely to reach any CCLs. For more information, read the Gemini 3.1 Pro Model Card.

Risks and Mitigations: Safety and responsibility was built into Gemini 3.1 Flash Image throughout the
training and deployment lifecycle, including pre-training, post-training, and product-level mitigations.
Mitigations include, but are not limited to:

    ●​   dataset filtering;
    ●​   conditional pre-training;
    ●​   supervised fine-tuning;
    ●​   reinforcement learning from human and critic feedback;
    ●​   safety policies and desiderata;
    ●​   product-level mitigations such as safety filtering.

                                                                                                           6