# Image-text-to-text Image-text-to-text models, also known as vision language models (VLMs), are language models that take an image input. These models can tackle various tasks, from visual question answering to image segmentation. This task shares many similarities with image-to-text, but with some overlapping use cases like image captioning. Image-to-text models only take image inputs and often accomplish a specific task, whereas VLMs take open-ended text and image inputs and are more generalist models. In this guide, we provide a brief overview of VLMs and show how to use them with Transformers for inference. To begin with, there are multiple types of VLMs: - base models used for fine-tuning - chat fine-tuned models for conversation - instruction fine-tuned models This guide focuses on inference with an instruction-tuned model. Let's begin installing the dependencies. ```bash pip install -q transformers accelerate pip install flash-attn --no-build-isolation ``` Let's initialize the model and the processor. ```python from transformers import AutoProcessor, AutoModelForImageTextToText from accelerate import Accelerator import torch device = Accelerator().device model = AutoModelForImageTextToText.from_pretrained( "Qwen/Qwen3-VL-4B-Instruct", dtype=torch.bfloat16, attn_implementation="flash_attention_2", ).to(device) processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-4B-Instruct") ``` This model has a [chat template](../chat_templating) that helps user parse chat outputs. Moreover, the model can also accept multiple images as input in a single conversation or message. We will now prepare the inputs. The image inputs look like the following. Structure your conversation as shown below for a single prompt with image and text inputs. ```python messages = [ { "role": "user", "content": [ {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats.png"}, {"type": "text", "text": "What do we see in this image?"}, ] } ] ``` Alternate between the `user` and `assistant` role to ground the model with prior context to generate better responses. ```python messages = [ { "role": "user", "content": [ {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats.png"}, {"type": "text", "text": "What do we see in this image?"}, ] }, { "role": "assistant", "content": [ {"type": "text", "text": "In this image we can see two cats on the nets."}, ] }, { "role": "user", "content": [ {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"}, {"type": "text", "text": "And how about this image?"}, ] }, ] ``` We will now call the processors' [apply_chat_template()](/docs/transformers/v5.8.0/en/main_classes/processors#transformers.ProcessorMixin.apply_chat_template) method to preprocess its output along with the image inputs. ```python inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(device) ``` We can now pass the preprocessed inputs to the model. ```python input_len = len(inputs.input_ids[0]) with torch.no_grad(): generated_ids = model.generate(**inputs, max_new_tokens=200) generated_texts = processor.batch_decode(generated_ids[:, input_len:], skip_special_tokens=True) print(generated_texts) ## ['In this image we can see flowers, plants and insect.'] ``` ## Pipeline The fastest way to get started is to use the [Pipeline](/docs/transformers/v5.8.0/en/main_classes/pipelines#transformers.Pipeline) API. Specify the `"image-text-to-text"` task and the model you want to use. ```python from transformers import pipeline pipe = pipeline("image-text-to-text", model="llava-hf/llava-interleave-qwen-0.5b-hf") ``` The example below uses chat templates to format the text inputs. ```python messages = [ { "role": "user", "content": [ { "type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg", }, {"type": "text", "text": "Describe this image."}, ], }, { "role": "assistant", "content": [ {"type": "text", "text": "There's a pink flower"}, ], }, ] ``` Pass the chat template formatted text and image to [Pipeline](/docs/transformers/v5.8.0/en/main_classes/pipelines#transformers.Pipeline) and set `return_full_text=False` to remove the input from the generated output. ```python outputs = pipe(text=messages, max_new_tokens=20, return_full_text=False) outputs[0]["generated_text"] # with a yellow center in the foreground. The flower is surrounded by red and white flowers with green stems ``` If you prefer, you can also load the images separately and pass them to the pipeline like so: ```python pipe = pipeline("image-text-to-text", model="HuggingFaceTB/SmolVLM-256M-Instruct") img_urls = [ "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats.png", "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg", ] images = [ Image.open(requests.get(img_urls[0], stream=True).raw), Image.open(requests.get(img_urls[1], stream=True).raw), ] messages = [ { "role": "user", "content": [ {"type": "image"}, {"type": "image"}, {"type": "text", "text": "What do you see in these images?"}, ], } ] outputs = pipe(text=messages, images=images, max_new_tokens=50, return_full_text=False) outputs[0]["generated_text"] " In the first image, there are two cats sitting on a plant. In the second image, there are flowers with a pinkish hue." ``` The images will still be included in the `"input_text"` field of the output: ```python outputs[0]['input_text'] """ [{'role': 'user', 'content': [{'type': 'image', 'image': }, {'type': 'image', 'image': }, {'type': 'text', 'text': 'What do you see in these images?'}]}]## Streaming """ ``` We can use [text streaming](./generation_strategies#streaming) for a better generation experience. Transformers supports streaming with the [TextStreamer](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.TextStreamer) or [TextIteratorStreamer](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.TextIteratorStreamer) classes. We will use the [TextIteratorStreamer](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.TextIteratorStreamer) with IDEFICS-8B. Assume we have an application that keeps chat history and takes in the new user input. We will preprocess the inputs as usual and initialize [TextIteratorStreamer](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.TextIteratorStreamer) to handle the generation in a separate thread. This allows you to stream the generated text tokens in real-time. Any generation arguments can be passed to [TextIteratorStreamer](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.TextIteratorStreamer). ```python import time from transformers import TextIteratorStreamer from threading import Thread def model_inference( user_prompt, chat_history, max_new_tokens, images ): user_prompt = { "role": "user", "content": [ {"type": "image"}, {"type": "text", "text": user_prompt}, ] } chat_history.append(user_prompt) streamer = TextIteratorStreamer( processor.tokenizer, skip_prompt=True, timeout=5.0, ) generation_args = { "max_new_tokens": max_new_tokens, "streamer": streamer, "do_sample": False } # add_generation_prompt=True makes model generate bot response prompt = processor.apply_chat_template(chat_history, add_generation_prompt=True) inputs = processor( text=prompt, images=images, return_tensors="pt", ).to(device) generation_args.update(inputs) thread = Thread( target=model.generate, kwargs=generation_args, ) thread.start() acc_text = "" for text_token in streamer: time.sleep(0.04) acc_text += text_token if acc_text.endswith(""): acc_text = acc_text[:-18] yield acc_text thread.join() ``` Now let's call the `model_inference` function we created and stream the values. ```python generator = model_inference( user_prompt="And what is in this image?", chat_history=messages[:2], max_new_tokens=100, images=images ) for value in generator: print(value) # In # In this # In this image ... ``` ## Fit models in smaller hardware VLMs are often large and need to be optimized to fit on smaller hardware. Transformers supports many model quantization libraries, and here we will only show int8 quantization with [Quanto](./quantization/quanto#quanto). int8 quantization offers memory improvements up to 75 percent (if all weights are quantized). However it is no free lunch, since 8-bit is not a CUDA-native precision, the weights are quantized back and forth on the fly, which adds up to latency. First, install dependencies. ```bash pip install -U optimum-quanto bitsandbytes ``` To quantize a model during loading, we need to first create [QuantoConfig](/docs/transformers/v5.8.0/en/main_classes/quantization#transformers.QuantoConfig). Then load the model as usual, but pass `quantization_config` during model initialization. ```python from transformers import AutoModelForImageTextToText, QuantoConfig model_id = "Qwen/Qwen3-VL-4B-Instruct" quantization_config = QuantoConfig(weights="int8") quantized_model = AutoModelForImageTextToText.from_pretrained( model_id, device_map="auto", quantization_config=quantization_config ) messages = [ { "role": "user", "content": [ {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats.png"}, {"type": "text", "text": "What do we see in this image?"}, ] }, ] inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device) input_len = len(inputs.input_ids[0]) with torch.no_grad(): generated_ids = model.generate(**inputs, cache_implementation="static", max_new_tokens=100) generated_texts = processor.batch_decode(generated_ids[:, input_len:], skip_special_tokens=True) print(generated_texts[0]) ## ['In this image, we see two tabby cats resting on a large, tangled pile of fishing nets. The nets are a mix of brown, orange, and red colors, with some blue and green ropes visible in the background. The cats appear relaxed and comfortable, nestled into the fibers of the nets. One cat is in the foreground, looking slightly to the side, while the other is positioned further back, looking directly at the camera. The scene suggests a coastal or fishing-related setting, possibly near'] ``` And that's it, we can use the model the same way with no changes. ## Iterative chatting with cache If you're building multimodal chat agents, you’re likely passing the same images or audio across turns. Caching lets you reuse their encoded representations instead of reprocessing them each time, cutting redundant computation. Like text-only models, multimodal models track dialogue history with a [chat template](./chat_templating). The key difference is that it applies the chat template in `"text"` format only and processes new multimodal content separately. If you apply the template with processing, the entire conversation gets reprocessed every turn because Jinja can't distinguish previously encoded inputs from new ones. The example below demonstrates this pattern. For text-only models, see the [iterative generation](../kv_cache#iterative-generation) guide. Here’s a simple Python example demonstrating this pattern. For an example with text-only models, refer to [this guide](../kv_cache#iterative-generation). ```python import torch from transformers import AutoProcessor, Gemma4ForConditionalGeneration, TextStreamer from transformers.cache_utils import DynamicCache from transformers.image_utils import load_image from transformers.audio_utils import load_audio model_id = 'google/gemma-4-E2B-it' processor = AutoProcessor.from_pretrained(model_id) model = Gemma4ForConditionalGeneration.from_pretrained(model_id, dtype=torch.float32, device_map='cpu') past_key_values = DynamicCache() streamer = TextStreamer(processor.tokenizer, skip_prompt=True, skip_special_tokens=True) ## Turn 1: multimodal input (text + image + audio) audio = load_audio("https://huggingface.co/datasets/Xenova/transformers.js-docs/resolve/main/jfk.wav") image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg") messages = [ { "role": "user", "content": [ {"type": "text", "text": "In detail, describe the following audio and image."}, {"type": "audio"}, {"type": "image"}, ], }, ] input_string = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) inputs = processor(text=input_string, images=[image], audio=[audio], return_tensors='pt') output = model.generate(**inputs, past_key_values=past_key_values, max_new_tokens=1024, do_sample=False, streamer=streamer) first_response = processor.decode(output[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True) ## Turn 2: text-only follow-up (reuses KV cache, no new media) messages.append({"role": "assistant", "content": [{"type": "text", "text": first_response}]}) cached_string = processor.apply_chat_template(messages, tokenize=False) messages.append({"role": "user", "content": [{"type": "text", "text": "Summarize the previous descriptions in one sentence."}]}) full_string = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) new_string = full_string[len(cached_string):] new_input_ids = processor.tokenizer(new_string, add_special_tokens=False, return_tensors='pt')['input_ids'] # Build and pass an attention mask only if batched input or there is padding # attention_mask = torch.ones(1, past_key_values.get_seq_length() + new_input_ids.shape[1], dtype=torch.long) output2 = model.generate(input_ids=new_input_ids, past_key_values=past_key_values, max_new_tokens=1024, do_sample=False, streamer=streamer) second_response = processor.decode(output2[0][new_input_ids.shape[1]:], skip_special_tokens=True) ## Turn 3: new image (must go through processor for image token expansion + encoding) image2 = load_image("https://huggingface.co/datasets/Xenova/transformers.js-docs/resolve/main/artemis.jpeg") messages.append({"role": "assistant", "content": [{"type": "text", "text": second_response}]}) cached_string2 = processor.apply_chat_template(messages, tokenize=False) messages.append({"role": "user", "content": [{"type": "text", "text": "Describe this image."}, {"type": "image"}]}) full_string2 = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) new_string2 = full_string2[len(cached_string2):] new_inputs2 = processor(text=new_string2, images=[image2], add_special_tokens=False, return_tensors='pt') # new_inputs2['attention_mask'] = torch.ones(1, past_key_values.get_seq_length() + new_inputs2['input_ids'].shape[1], dtype=torch.long) new_inputs2['past_key_values'] = past_key_values output3 = model.generate(**new_inputs2, max_new_tokens=1024, do_sample=False, streamer=streamer) ``` ## Further Reading Here are some more resources for the image-text-to-text task. - [Image-text-to-text task page](https://huggingface.co/tasks/image-text-to-text) covers model types, use cases, datasets, and more. - [Vision Language Models Explained](https://huggingface.co/blog/vlms) is a blog post that covers everything about vision language models and supervised fine-tuning using [TRL](https://huggingface.co/docs/trl/en/index). - [Learn how to fine-tune vision language models using TRL](https://huggingface.co/blog/trl-vlm-alignment)