TechPrufer
Get started
Google
Google logoGoogle · Feature trail

Diffusion Models for Text Generation

27 posts · Dec 2022 – Jun 2026

What this trail covers

Google has introduced Imagen 4, its latest text-to-image model, available via the Gemini API and Google AI Studio. This model family includes Imagen 4 and Imagen 4 Ultra, offering improved text rendering and prompt adherence. Images generated by these models are watermarked with SynthID for trust and transparency. Developers can access these models through documentation and cookbooks, with plans for broader availability and additional billing tiers. The Gemini API now supports Imagen 4, offering improved quality and precision for text-to-image generation.

All Google trails
  1. DiffusionGemma: The Developer Guide- Google Developers BlogThis post introduces DiffusionGemma, an experimental model built on the Gemma 4 backbone, which offers compute-bound parallel generation, bidirectional context and self-correction, and developer-friendly sizes. It details the architecture, including uniform state diffusion and block autoregressive diffusion for variable length generation. The post showcases its ability to solve Sudoku puzzles through parallel denoising and error correction via re-noising. It also provides guidance on serving DiffusionGemma using vLLM and links to resources for downloading weights, learning more, and adapting/fine-tuning the model.
  2. Introducing Veo 3.1 and new creative capabilities in the Gemini API- Google Developers BlogIntroduces Veo 3.1 and Veo 3.1 Fast, updated video generation models available via the Gemini API, Google AI Studio, Vertex AI, Gemini app, and Flow. New capabilities include richer native audio, improved cinematic style control, enhanced image-to-video generation using reference images for character/style consistency, scene extension for creating longer videos by generating clips based on the previous video's final second, and first/last frame control for generating transitions between two images with accompanying audio. Code examples are provided for using reference images, scene extension, and first/last frame control with the Gemini API.
  3. Own your AI: Learn how to fine-tune Gemma 3 270M and run it on-device- Google Developers BlogThis post details the process of fine-tuning Gemma 3 270M for a specific task (text to emoji translation) using QLoRA for parameter-efficient fine-tuning. It covers quantizing the model for on-device inference (reducing size to under 300MB) and deploying it in a web app using MediaPipe or Transformers.js, leveraging WebGPU for client-side LLM inference. It provides links to Colab notebooks for fine-tuning and conversion, and a GitHub repository for the example web app.
  4. Gemini 2.5 Flash Image now ready for production with new aspect ratios- Google Developers BlogThe Gemini 2.5 Flash Image model is now generally available for production, with new features including support for 10 different aspect ratios (landscape, square, portrait, and flexible) and the ability to specify image-only output. This model empowers users to seamlessly blend multiple images, maintain consistent characters for richer storytelling, perform targeted edits with natural language, and leverage Gemini's extensive world knowledge for image generation and modification. Examples of its use are provided by Cartwheel and Volley, showcasing its capabilities in character control, consistency, and real-time applications. The post also includes a Python code sample for generating content with the model and details pricing information.
  5. Unlocking Multi-Spectral Data with Gemini- Google Developers BlogThis post details a novel technique for using Gemini models to analyze multi-spectral imagery by mapping invisible spectral bands (NIR, SWIR) to RGB channels, creating 'false-color composite' images. This allows Gemini's reasoning engine to interpret complex remote sensing data without requiring custom-trained models. The process involves selecting three bands, normalizing and mapping them to R, G, B, and then providing context in the prompt. Examples demonstrate Gemini's improved accuracy in classifying land cover types (e.g., River vs. Forest) when provided with these multi-spectral inputs. A Colab notebook is provided for practical application.
  7. Introducing EmbeddingGemma: The Best-in-Class Open Model for On-Device Embeddings- Google Developers BlogIntroduced EmbeddingGemma, a new open embedding model based on the Gemma 3 architecture, optimized for on-device AI. It offers best-in-class performance for its size (308M parameters), supporting RAG and semantic search directly on hardware. Key features include multilingual capabilities (100+ languages), low RAM usage (<200MB with quantization), customizable output dimensions via Matryoshka Representation Learning, and fast inference times (<15ms on EdgeTPU). It integrates with popular tools and frameworks, enabling offline and privacy-centric applications. The post also contrasts it with the Gemini Embedding model for server-side applications and provides links to download models, documentation, and examples.
  8. How to prompt Gemini 2.5 Flash Image Generation for the best results- Google Developers BlogThis post provides a guide on how to effectively prompt Gemini 2.5 Flash Image Generation for optimal results. It details techniques for text-to-image generation, including photorealistic scenes, stylized illustrations, accurate text rendering, product mockups, minimalist designs, and sequential art. It also covers image editing capabilities such as adding/removing elements, inpainting specific areas, style transfer, and advanced composition by combining multiple images. The post emphasizes descriptive prompting and provides templates and examples for each use case.
  9. Introducing Gemini 2.5 Flash Image, our state-of-the-art image model- Google Developers BlogIntroduces Gemini 2.5 Flash Image (nano-banana), a state-of-the-art image generation and editing model. Key capabilities include: blending multiple images, maintaining character consistency across prompts/edits, targeted transformations via natural language prompts for precise local edits (e.g., blurring backgrounds, removing objects, adding color), and leveraging Gemini's world knowledge for semantically richer image generation. It also supports multi-image fusion. The model is available via Gemini API, Google AI Studio, and Vertex AI, with pricing at $0.039 per image. Demo apps are provided in Google AI Studio. Images are watermarked with SynthID. Future improvements will focus on long-form text rendering, character consistency, and factual representation.
  10. Announcing Imagen 4 Fast and the general availability of the Imagen 4 family in the Gemini API- Google Developers BlogAnnounces the general availability of Imagen 4, Imagen 4 Fast, and Imagen 4 Ultra in the Gemini API and Google AI Studio. Highlights Imagen 4 Fast's speed and cost ($0.02/image), Imagen 4's improved text rendering, and Imagen 4 Ultra's detail adherence. Notes support for up to 2K resolution for Imagen 4 and Imagen 4 Ultra. Mentions imperceptible watermarking with SynthID.
  11. T5Gemma: A new collection of encoder-decoder Gemma models- Google Developers BlogIntroduces T5Gemma, a new collection of encoder-decoder LLMs derived from decoder-only Gemma models via an adaptation technique. Details the model adaptation process, which involves initializing encoder-decoder parameters with weights from pretrained decoder-only models and further pre-training. Highlights T5Gemma's performance advantages in quality-inference efficiency across benchmarks like SuperGLUE and GSM8K, demonstrating comparable or better results than decoder-only Gemma models. Discusses the flexibility of 'unbalanced' encoder-decoder configurations (e.g., 9B encoder with 2B decoder) for task-specific trade-offs. Releases T5Gemma checkpoints in various sizes (Small, Base, Large, XL, 2B, 9B) and variants (pretrained, instruction-tuned), including an unbalanced 9B-2B model, and models trained with PrefixLM or UL2 objectives.
  12. Imagen 4 is now available in the Gemini API and Google AI Studio- Google Developers BlogIntroduces Imagen 4 and Imagen 4 Ultra, Google's latest text-to-image models, available in paid preview via the Gemini API and for limited testing in Google AI Studio. Highlights improved text rendering and prompt adherence compared to prior models. Details pricing for Imagen 4 ($0.04/image) and Imagen 4 Ultra ($0.06/image), with plans for additional billing tiers and rate limit increase requests. Provides examples of generated images and links to documentation and cookbooks. Mentions SynthID watermarking for trust and transparency.
  13. Multilingual innovation in LLMs: How open models help unlock global communication- Google Developers BlogThis post highlights the Kaggle competition 'Unlock Global Communication with Gemma' where developers adapted Gemma LLMs for diverse cultural and linguistic contexts, addressing performance gaps in lower-resourced languages. It showcases specific projects like 'Gemma 2 Swahili', 'Kyara: Retrieval Augmentation for LLM Fine-Tuning', 'ArGemma: Fine-Tuning Gemma for Arabic', 'Post-Training Gemma for Italian and beyond', 'Ancient Chinese Expert: Gemma 2>ChatGPT', 'Lyric-Gemma 2: One Song, Different Stories', 'Fine-tuning Gemma 2 JPN for Yomigana', 'Mathematical Minds: Fine-tuning Gemma 2 for Hindi', 'Gemma-2-9b-kk-it: Learning to translate Kazakh', and 'THEODEN: The Old English Gemma'. These projects demonstrate parameter-efficient fine-tuning, retrieval augmentation, handling of historical languages, and adaptation for specific tasks like translation, summarization, and mathematical reasoning. It also mentions Gemma 3's pretrained support for over 140 languages.
  14. Create and edit images with Gemini 2.0 in preview- Google Developers BlogIntroduces Gemini 2.0 Flash for image generation and editing via Gemini API in Google AI Studio and Vertex AI. Highlights improved visual quality, text rendering, and reduced filter block rates. Demonstrates functionalities like product recontextualization, collaborative editing, conversational editing, dynamic SKU creation, and ideation partnership. Provides a Python code example for generating content with images.
  15. Gemma explained: What’s new in Gemma 3- Google Developers BlogThis post introduces Gemma 3, detailing its new vision-language capabilities enabled by a SigLIP encoder and a "Pan&Scan" algorithm for image processing. It explains the use of "soft tokens" to reduce inference resource requirements. Architectural improvements include 5-to-1 interleaved attention for better context handling and QK-norm for improved accuracy and speed. The post highlights Gemma 3's extended context length support (up to 128k tokens) and its use of bidirectional attention for image inputs, contrasting it with PaliGemma's autoregressive approach. A new tokenizer with a 262k vocabulary size is also introduced for enhanced multilingual support.
  16. Bring your ideas to life: Veo 2 video generation available for developers- Google Developers BlogThis post announces the availability of Veo 2, a state-of-the-art video generation model, within the Gemini API and Google AI Studio. It details Veo 2's core capabilities: text-to-video (t2v) and image-to-video (i2v) generation, producing eight-second video clips from text, image, or combined prompts. The post provides guidance on crafting effective prompts with examples, a Python code snippet for API integration, and showcases real-world applications through AlphaWave and Trakto Studio. It highlights the model's ability to simulate real-world physics and capture diverse visual styles, and encourages developers to experiment in Google AI Studio, explore the Colab Notebook, and consult the API documentation.
  17. Experiment with Gemini 2.0 Flash native image generation- Google Developers BlogThis post announces the availability of native image generation in Gemini 2.0 Flash for developer experimentation. It highlights the model's multimodal capabilities for text and image storytelling, conversational image editing, world understanding for realistic imagery, and improved text rendering. The post provides examples and a Python code snippet demonstrating how to use the experimental version of Gemini 2.0 Flash for image generation.
  18. Safer and Multimodal: Responsible AI with Gemma- Google Developers BlogIntroduces ShieldGemma 2, a 4B parameter safety classifier model built on Gemma 3, extending safety capabilities to image analysis for multimodal AI. It can be used as an input filter for vision language models or an output filter for image generation systems, handling both synthetic and natural images. The model is trained on curated datasets and instruction-tuned for performance, with a focus on detecting sexually explicit, dangerous, and violent content. It offers flexibility for fine-tuning, versatility across Gemma 3 supporting frameworks, and an open, collaborative approach.
  19. Imagen 3 arrives in the Gemini API- Google Developers BlogThis post announces the availability of Imagen 3, Google's latest image generation model, through the Gemini API. It highlights Imagen 3's improved prompt following, artifact-free image generation across various styles, and state-of-the-art performance on benchmarks. The post also details pricing, the inclusion of SynthID watermarking for combating misinformation, and provides a Python code snippet for generating images. It emphasizes the expansion of generative media models into the Gemini API.
  20. Introducing PaliGemma 2: Powerful Vision-Language Models, Simple Fine-Tuning- Google Developers BlogIntroduces PaliGemma 2, an evolution of vision-language models built on Gemma 2. Highlights scalable performance with multiple model sizes and resolutions, long captioning capabilities, and leading performance on specific recognition tasks (chemical formulas, music scores, spatial reasoning, chest X-rays). Positions PaliGemma 2 as a drop-in replacement for existing PaliGemma users and emphasizes straightforward fine-tuning.
  21. Announcing Build with Google AI release 3: A Season of Gemma!- Google Developers BlogThis post announces Season 3 of the 'Build with Google AI' series, focusing on practical applications of Google's Gemma open generative AI models. It highlights three specific projects: a Native Language AI Assistant, a Business Email AI Assistant, and a Personal AI Code Assistant (upgraded to work with Gemma and a VS Code extension). All projects are provided with open-source code and guidance from Google AI experts. It also directs users to Gemma developer documentation and the Gemma cookbook repository for further resources.
  22. Gemma explained: RecurrentGemma architecture- Google Developers BlogThis post introduces the RecurrentGemma architecture, detailing its hybrid approach combining gated linear recurrences with local sliding window attention. It explains the architectural components like RecurrentGemmaDecoderLayer, RecurrentGemmaRecurrentBlock, RecurrentGemmaSdpaAttention, and RecurrentGemmaMlp. Key parameters such as non-embedding and embedding parameters, model width, RNN width, MLP expansion factor, and local attention window size are discussed. The post also provides a Python code snippet for the RecurrentGemmaForCausalLM model and elaborates on the embed_tokens layer and the structure of the 26 decoder layers, including the specific implementation of residual blocks with recurrent blocks and local MQA blocks.
  23. Gemma explained: What’s new in Gemma 2- Google Developers BlogThis post introduces Gemma 2, detailing its architectural innovations including alternating local and global attention, logit soft-capping, RMSNorm for pre and post-normalization, and Grouped-Query Attention (GQA). It provides specific implementation details for the Gemma 27B model's self-attention mechanism and MLP layers, including the number of heads and head sizes for GQA across different model sizes (27B, 9B, 2B). Key findings from the technical report are summarized, highlighting the benefits of knowledge distillation, GQA over MHA, and deeper vs. wider model architectures.
  24. Gemma explained: An overview of Gemma model family architectures- Google Developers BlogThis post introduces the Gemma family of models, detailing their architectures and design choices. It explains the decoder-only transformer architecture, core parameters like d_model, layers, feedforward hidden dimensions, num_heads, num_kv_heads (distinguishing MHA in 7B from MQA in 2B), head_size, and vocab_size. It provides a detailed breakdown of the Gemma 7B model's architecture, including its embedding layer, decoder layers, self-attention mechanism (with specific details on projections, heads, and rotary embeddings), MLP with GeGLU activation, and language modeling head. It also highlights the use of SentencePiece for tokenization and the context length of 8192 tokens.
  25. Smaller, Safer, More Transparent: Advancing Responsible AI with Gemma- Google Developers BlogIntroduces Gemma 2 2B, a new 2 billion parameter model optimized for on-device deployment and efficiency, outperforming GPT-3.5 on the Chatbot Arena. Details its integration with NVIDIA TensorRT-LLM and availability on various hardware. Introduces ShieldGemma, a suite of safety content classifier models built on Gemma 2 to filter harmful AI inputs/outputs (hate speech, harassment, sexually explicit, dangerous content), offering flexible sizes (2B, 9B, 27B) and SOTA performance. Introduces Gemma Scope, a model interpretability tool using sparse autoencoders (SAEs) to analyze Gemma 2 2B and 9B models, with over 400 SAEs available and interactive demos on Neuronpedia.
  26. 3 fun Gemma project ideas- Google Developers BlogThis post introduces three project ideas that leverage Gemma models: translating old Korean literature, game design brainstorming, and a "magic of Santa's mailbox" project. For the Korean translation project, it details the use of fine-tuning with historical datasets and SentencePiece for tokenization, and provides code examples for saving and uploading fine-tuned models. For game design, it shows how Gemma can generate game concepts, characters, and stories based on prompts. The "magic of Santa's mailbox" project is outlined as using Gemma on a Raspberry Pi. Expansion ideas for style transfer and dialect translation are also presented.
  27. Fine-tuning Gemma 2 with Keras - and an update from Hugging Face- Google Developers BlogThis post details the fine-tuning of Gemma 2 models using Keras and JAX, focusing on distributed fine-tuning with Model Parallelism and LoRA. It introduces the `keras.distribution.ModelParallel` API for specifying weight sharding configurations and demonstrates how to use `keras.distribution.LayoutMap` for layer-by-layer weight partitioning. It also showcases the use of LoRA for efficient fine-tuning by freezing model weights and replacing them with low-rank adapters. Furthermore, it announces enhanced Hugging Face integration allowing KerasNLP to load fine-tuned weights from Hugging Face for supported models, including Gemma and Llama3, and highlights the availability of PaliGemma for vision-language tasks.
  28. How it's Made: GenType Alphabet Creator- Google Developers BlogThis post details the creation of GenType, a tool that uses the Imagen 2 API to generate an entire alphabet from a single user input. It explains the prompt recipe used ('The letter "{letter}" made out of {prompt}') and the automation process of sending 26 individual requests for each letter. It also provides guidance on crafting effective prompts by specifying foreground, background, and style for consistent results. The post highlights the flexibility of the tool for various creative use cases and encourages submissions of similar AI experiments to labs.google.
  29. Introducing PaliGemma, Gemma 2, and an Upgraded Responsible AI Toolkit- Google Developers BlogThis post announces the upcoming launch of Gemma 2, detailing its new architecture designed for breakthrough performance and efficiency, with a 27B parameter model comparable to Llama 3 70B at less than half the size. It also introduces PaliGemma, an open vision-language model (VLM) built on SigLIP and Gemma, available with pretrained and fine-tuned checkpoints for various vision-language tasks. The Responsible Generative AI Toolkit is expanded with the open-source LLM Comparator for model evaluation.
  30. Gemma Family Expands with Models Tailored for Developers and Researchers- Google Developers BlogIntroduces CodeGemma, a 7B pretrained and instruction-tuned variant, and a 2B pretrained variant, specialized for code completion and generation tasks. Also introduces RecurrentGemma, an efficiency-optimized architecture leveraging recurrent neural networks and local attention for improved memory efficiency and higher throughput, showcasing a non-transformer model. Updates Gemma 1.1 with performance improvements and bug fixes. Details compatibility with various frameworks and hardware.
  31. Introducing Gemma models in Keras- Google Developers BlogThis post announces the integration of Gemma models into the KerasNLP collection, leveraging Keras 3's multi-backend support (JAX, PyTorch, TensorFlow). It introduces a new LoRA API for parameter-efficient fine-tuning, significantly reducing trainable parameters. Additionally, it details model-parallel training capabilities using Keras's distribution API, enabling distributed training across multiple GPUs/TPUs by sharding model weights. Code examples are provided for getting started, LoRA fine-tuning, and distributed training.
  32. Make with MakerSuite Part 2: Tuning LLMs- Google Developers BlogThis post introduces Parameter Efficient Tuning (PET) within MakerSuite, a technique that allows developers to customize Large Language Models (LLMs) with hundreds of data points in minutes. It contrasts PET with traditional fine-tuning, highlighting PET's reduced complexity, data requirements, and computational cost. The post details the workflow for tuning models in MakerSuite, including creating a tuned model, selecting data (from saved prompts, Google Sheets, or CSV), viewing tuning progress, and running the newly tuned model.
  33. Improving Video Voice Dubbing Through Deep Learning- Google Developers BlogThis post details the use of deep learning for video voice dubbing, employing techniques like sequence-to-sequence models with attention mechanisms to align audio and video streams, and using diffusion models to generate natural-sounding speech that matches the original speaker's prosody and emotion. The system leverages large language models for text-to-speech synthesis and fine-tuning for specific voice characteristics.