Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Pruna AI / P-Image

P-Image

Pruna text-to-image model with custom dimensions and optional style or subject adapters.

Image generation
availableClosed weightsImage generation
Official documentation (opens in a new tab)Reviewed 15 Oct 2026
7 of 22 specification fields collected
Context windowNot collectedAwaiting source review
Maximum outputNot collectedAwaiting source review
Input billingSee ratesPruna API
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Maximum input tokens
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Maximum output
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Maximum prompt characters
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Reasoning
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Tool calling
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Structured output
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Streaming
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Batch processing
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Context caching
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Knowledge cutoff
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Release date
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Parameters
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Architecture
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Languages
Not collected

Not established for this exact hosted image model. No language-model token budget or capability is inferred from its JSON API.

Input modalities
TextSource for Input modalities (opens in a new tab)
Output modalities
ImageSource for Output modalities (opens in a new tab)
Fine-tuning
SupportedSource for Fine-tuning (opens in a new tab)

Hosted LoRA adapter training only; full base-weight fine-tuning is not established.

Custom model training
Separate LoRA trainer produces downloadable adapter weights. Adapter inference uses the corresponding LoRA route, not a new base model.Source for Custom model training (opens in a new tab)
Aspect ratios
1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, customSource for Aspect ratios (opens in a new tab)
Usage metering
Optional adapter training: $1.80/1,000 steps, separate from image inference.Source for Usage metering (opens in a new tab)
Supported image sizes
Custom width and height: 256–1440 px, each a multiple of 16.Source for Supported image sizes (opens in a new tab)

Custom dimensions apply only with aspect_ratio=custom.

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Pruna API

Provider source (opens in a new tab)
API model ID
p-image
Access
Native hosted inference requires an account and API key. Availability is documented, not tested with a generation request.
Regions
Not established
API endpoint URL
POST https://api.pruna.ai/v1/predictionsSource for API endpoint URL (opens in a new tab)
Authentication
apikey header; Model header selects p-image. JSON body wraps parameters in input.Source for Authentication (opens in a new tab)
Required request fields
promptSource for Required request fields (opens in a new tab)

Required properties in OpenAPI 0.6.0. Guide recommendations and image constraints are recorded separately.

Model controls
seed, width, height, prompt, lora_scale, aspect_ratio, hf_api_token, lora_weights, prompt_upsampling, disable_safety_checkerSource for Model controls (opens in a new tab)
Request quotas
500 requests/minuteSource for Request quotas (opens in a new tab)

Published model quota; concurrency and account-specific adjustments are not specified.

Asynchronous generation
Asynchronous by default. Try-Sync=true waits up to 60 seconds; the API can return an asynchronous prediction if not completed.Source for Asynchronous generation (opens in a new tab)
Result retrieval
Poll GET /v1/predictions/status/{id} with the API key; download the returned result URL.Source for Result retrieval (opens in a new tab)
Pricing coverage
Per generated image. No language-token tariff or cost-per-question conversion.Source for Pricing coverage (opens in a new tab)
Aspect ratios
1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, customSource for Aspect ratios (opens in a new tab)
Seed control
Optional integer seed. Numeric range is not specified in this OpenAPI revision.Source for Seed control (opens in a new tab)
Content screening controls
disable_safety_checker defaults to false.Source for Content screening controls (opens in a new tab)
Resolution controls
Default aspect ratio 16:9. For custom dimensions, width and height each 256–1440 px; description requires multiples of 16.Source for Resolution controls (opens in a new tab)

The multiple-of-16 condition is in the description, not a multipleOf schema keyword.

Prompt expansion
prompt_upsampling defaults to false.Source for Prompt expansion (opens in a new tab)

The introductory prose mentions automatic enhancement; the explicit request-field default is preserved.

Adapter controls
lora_weights accepts a Hugging Face repository URL with an optional .safetensors filename. lora_scale ranges from -1 to 3; default 0.5. Private repositories use hf_api_token.Source for Adapter controls (opens in a new tab)
Adapter compatibility
Use adapters trained by p-image-trainer. The image-generation and editing trainers are not interchangeable.Source for Adapter compatibility (opens in a new tab)
Output formats
Not collected

Examples use image download URLs but do not establish an exhaustive output-format contract for this route.

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
CategoryUsageRateUnit
Image generationImage generation1$0.005image

Rate conditions

  1. Native Pruna route. Adapter training is charged separately. Replicate prices, discounts and enterprise terms are not inferred.
Pricing source (opens in a new tab)

Data & enterprise terms

API terms retain customer ownership of inputs and outputs and grant Pruna processing rights to provide the service. Telemetry supports operations, security and improvement. Reviewed terms do not establish a universal retention period or a zero-training guarantee.

Read the service policy (opens in a new tab)

Pruna API

Provider source (opens in a new tab)
API model ID
p-image-lora
Access
Native hosted inference requires an account and API key. Availability is documented, not tested with a generation request.
Regions
Not established
API endpoint URL
POST https://api.pruna.ai/v1/predictionsSource for API endpoint URL (opens in a new tab)
Authentication
apikey header; Model header selects p-image-lora. JSON body wraps parameters in input.Source for Authentication (opens in a new tab)
Required request fields
promptSource for Required request fields (opens in a new tab)

Required properties in OpenAPI 0.6.0. Guide recommendations and image constraints are recorded separately.

Model controls
seed, width, height, prompt, lora_scale, aspect_ratio, hf_api_token, lora_weights, prompt_upsampling, disable_safety_checkerSource for Model controls (opens in a new tab)
Request quotas
500 requests/minuteSource for Request quotas (opens in a new tab)

Published model quota; concurrency and account-specific adjustments are not specified.

Asynchronous generation
Asynchronous by default. Try-Sync=true waits up to 60 seconds; the API can return an asynchronous prediction if not completed.Source for Asynchronous generation (opens in a new tab)
Result retrieval
Poll GET /v1/predictions/status/{id} with the API key; download the returned result URL.Source for Result retrieval (opens in a new tab)
Pricing coverage
Per generated image. No language-token tariff or cost-per-question conversion.Source for Pricing coverage (opens in a new tab)
Aspect ratios
1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, customSource for Aspect ratios (opens in a new tab)
Seed control
Optional integer seed. Numeric range is not specified in this OpenAPI revision.Source for Seed control (opens in a new tab)
Content screening controls
disable_safety_checker defaults to false.Source for Content screening controls (opens in a new tab)
Resolution controls
Default aspect ratio 16:9. For custom dimensions, width and height each 256–1440 px; description requires multiples of 16.Source for Resolution controls (opens in a new tab)

The multiple-of-16 condition is in the description, not a multipleOf schema keyword.

Prompt expansion
prompt_upsampling defaults to false.Source for Prompt expansion (opens in a new tab)

The introductory prose mentions automatic enhancement; the explicit request-field default is preserved.

Adapter controls
lora_weights accepts a Hugging Face repository URL with an optional .safetensors filename. lora_scale ranges from -1 to 3; default 0.5. Private repositories use hf_api_token.Source for Adapter controls (opens in a new tab)
Adapter compatibility
Use adapters trained by p-image-trainer. The image-generation and editing trainers are not interchangeable.Source for Adapter compatibility (opens in a new tab)
Output formats
Not collected

Examples use image download URLs but do not establish an exhaustive output-format contract for this route.

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
CategoryUsageRateUnit
Image generationImage generation1$0.005image

Rate conditions

  1. Native Pruna route. Adapter training is charged separately. Replicate prices, discounts and enterprise terms are not inferred.
Pricing source (opens in a new tab)

Data & enterprise terms

API terms retain customer ownership of inputs and outputs and grant Pruna processing rights to provide the service. Telemetry supports operations, security and improvement. Reviewed terms do not establish a universal retention period or a zero-training guarantee.

Read the service policy (opens in a new tab)

Pruna API (LoRA training)

Provider source (opens in a new tab)
API model ID
p-image-trainer
Access
Account and API key required. This is adapter training, not inference; no training job was submitted.
Regions
Not established
API endpoint URL
POST https://api.pruna.ai/v1/predictionsSource for API endpoint URL (opens in a new tab)
Authentication
apikey header; Model header selects p-image-trainer. JSON body wraps parameters in input.Source for Authentication (opens in a new tab)
Required request fields
image_dataSource for Required request fields (opens in a new tab)

Required properties in OpenAPI 0.6.0. Guide recommendations and image constraints are recorded separately.

Model controls
steps, image_data, learning_rate, training_type, default_captionSource for Model controls (opens in a new tab)
Request quotas
5 requests/minuteSource for Request quotas (opens in a new tab)

Published model quota; concurrency and account-specific adjustments are not specified.

Asynchronous generation
Asynchronous only. Do not use Try-Sync. Poll the prediction status URL.Source for Asynchronous generation (opens in a new tab)
Result retrieval
Poll GET /v1/predictions/status/{id} with the API key; download the returned result URL.Source for Result retrieval (opens in a new tab)
Pricing coverage
Adapter training charged per 1,000 steps, separately from inference.Source for Pricing coverage (opens in a new tab)
Training controls
steps: 100–5000, multiples of 100, default 1000. learning_rate: 0.00001–0.01, default 0.0001. training_type: content, style or balanced; default balanced.Source for Training controls (opens in a new tab)
Training dataset
ZIP URL containing images; the guide recommends at least 10. Matching image-name .txt files supply captions.Source for Training dataset (opens in a new tab)
Reference guidance
Supply default_caption where a caption file is missing; otherwise training fails. Dataset byte and image-resolution limits are not established.Source for Reference guidance (opens in a new tab)
Output formats
ZIP containing .safetensors adapter weights, not generated images or downloadable base weights.Source for Output formats (opens in a new tab)
Data retention
Training output URL expires approximately 30 minutes after completion; download promptly. This is delivery-URL expiry, not a general data-retention policy.Source for Data retention (opens in a new tab)
Adapter compatibility
Use weights with p-image-lora. P-Image and P-Image-Edit trainers are not interchangeable.Source for Adapter compatibility (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
CategoryUsageRateUnit
LoRA trainingAdapter training1$1.801,000 training steps

Rate conditions

  1. Optional adapter training. Independent of per-image inference; minimum and increment are 100 steps. Billing rounding is not established.
Pricing source (opens in a new tab)

Data & enterprise terms

API terms retain customer ownership of inputs and outputs and grant Pruna processing rights to provide the service. Telemetry supports operations, security and improvement. Reviewed terms do not establish a universal retention period or a zero-training guarantee.

Read the service policy (opens in a new tab)

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Pruna API

p-image
Not verified

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

How it works
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Minimum reusable input
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Lifetime & refresh
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

How to enable it
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Usage fields
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Billing rules
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Limits & data boundaries
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

Pruna API

p-image-lora
Not verified

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

How it works
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Minimum reusable input
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Lifetime & refresh
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

How to enable it
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Usage fields
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Billing rules
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Limits & data boundaries
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

Pruna API (LoRA training)

p-image-trainer
Not verified

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

How it works
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Minimum reusable input
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Lifetime & refresh
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

How to enable it
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Usage fields
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Billing rules
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

Limits & data boundaries
Not collected

Prompt-cache support and discounts are not established. Reusing uploaded images or adapter weights does not establish inference prompt caching.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Closed-weight classification is inferred: Pruna names and operates this hosted model, retains model rights under its API terms, and the reviewed public inventory exposes adapters rather than a matching base checkpoint. This does not establish an original architecture or exclude third-party ancestry. The Apache-licensed optimization library and public LoRAs are not base-model weight releases.

Classification evidence (opens in a new tab)

Sources reviewed