Fix sticker generation for newer GPT image models

Sticker generation was failing when using newer GPT image models
(gpt-image-1, gpt-image-1-mini, gpt-image-1.5). The issue occurred
because stickers requested 256x256 size, but these models only support
1024x1024, 1536x1024, 1024x1536, and auto.

To reproduce, send `!bai sticker Something` to an agent configured
with a GPT image model. The error was:

  invalid_request_error: Invalid value: '256x256'. Supported values
  are: '1024x1024', '1024x1536', '1536x1024', and 'auto'. (param: size)
  (code: invalid_value)

The fix replaces the hardcoded 256x256 size override with a
`smallest_size_possible` flag, letting each provider determine the
appropriate sticker size based on the model being used.

The `openai_compat` provider still defaults to requesting 256x256 in all cases
(regardless of model name).

src/agent/provider/entity/image.rs

@@ -2,7 +2,7 @@ use mxlink::mime;
 
 #[derive(Default)]
 pub struct ImageGenerationParams {
-    pub size_override: Option<String>,
+    pub smallest_size_possible: bool,
 
     pub cheaper_model_switching_allowed: bool,
 
@@ -10,8 +10,8 @@ pub struct ImageGenerationParams {
 }
 
 impl ImageGenerationParams {
-    pub fn with_size_override(mut self, value: Option<String>) -> Self {
+    pub fn with_smallest_size_possible(mut self, value: bool) -> Self {
-        self.size_override = value;
+        self.smallest_size_possible = value;
         self
     }
 

src/agent/provider/openai/controller.rs

@@ -31,12 +31,9 @@ use crate::{
 use crate::{
     agent::{
         AgentPurpose,
-        provider::{
+        provider::entity::{
-            entity::{
+            ImageEditResult, ImageGenerationResult, ImageSource, PingResult,
-                ImageEditResult, ImageGenerationResult, ImageSource, PingResult,
+            TextToSpeechParams, TextToSpeechResult,
-                TextToSpeechParams, TextToSpeechResult,
-            },
-            openai::utils::convert_string_to_enum,
         },
     },
     strings,
@@ -312,10 +309,11 @@ impl ControllerTrait for Controller {
             image_generation_config.quality.clone()
         };
 
-        let size = params
+        let size = if params.smallest_size_possible {
-            .size_override
+            Some(get_sticker_size(&model))
-            .map(|s| convert_string_to_enum::<async_openai::types::images::ImageSize>(&s).unwrap())
+        } else {
-            .or(image_generation_config.size);
+            image_generation_config.size
+        };
 
         let response_format = match model.clone() {
             ImageModel::DallE2 => Some(ImageResponseFormat::B64Json),
@@ -638,3 +636,17 @@ fn audio_mime_type_to_file_name(mime_type: &mxlink::mime::Mime) -> Option<String
 
     Some(format!("audio.{}", file_extension))
 }
+
+/// Returns the smallest supported size for stickers based on what the image model supports.
+fn get_sticker_size(model: &ImageModel) -> async_openai::types::images::ImageSize {
+    use async_openai::types::images::ImageSize;
+
+    match model {
+        ImageModel::DallE2 => ImageSize::S256x256,
+        ImageModel::DallE3 => ImageSize::S1024x1024,
+        ImageModel::GptImage1 => ImageSize::S1024x1024,
+        ImageModel::GptImage1Mini => ImageSize::S1024x1024,
+        ImageModel::GptImage1dot5 => ImageSize::S1024x1024,
+        ImageModel::Other(_) => ImageSize::S1024x1024,
+    }
+}

src/agent/provider/openai/utils.rs

@@ -1,8 +1,6 @@
-use async_openai::types::{
+use async_openai::types::responses::{
-    responses::{
+    EasyInputContent, EasyInputMessage, ImageDetail, InputContent, InputImageContent, InputItem,
-        EasyInputContent, EasyInputMessage, ImageDetail, InputContent, InputImageContent,
+    InputParam, MessageType, Role,
-        InputItem, InputParam, MessageType, Role,
-    },
 };
 
 use crate::conversation::llm::{
@@ -48,19 +46,3 @@ pub fn convert_llm_messages_to_openai_response_input(
 
     InputParam::Items(items)
 }
-
-pub(super) fn convert_string_to_enum<T>(value: &str) -> Result<T, String>
-where
-    T: serde::de::DeserializeOwned,
-{
-    // This is a hacky way to construct an enum from the string we have.
-    let enum_result: serde_json::Result<T> = serde_json::from_str(&format!("\"{}\"", value));
-    match enum_result {
-        Ok(enum_result) => Ok(enum_result),
-        Err(err) => {
-            tracing::debug!(?err, "Failed to parse into enum");
-
-            Err(format!("The value ({}) is not supported.", value))
-        }
-    }
-}

src/agent/provider/openai_compat/controller.rs

@@ -3,6 +3,8 @@ use etke_openai_api_rust::chat::{ChatApi, ChatBody};
 use etke_openai_api_rust::images::{ImagesApi, ImagesBody};
 use etke_openai_api_rust::{Auth, Message, OpenAI};
 
+const SMALLEST_IMAGE_SIZE: &str = "256x256";
+
 use super::super::ControllerTrait;
 use crate::utils::base64::base64_decode;
 use crate::{
@@ -303,9 +305,11 @@ impl ControllerTrait for Controller {
         // when they span multiple lines.
         let prompt = prompt.replace("\n", " ");
 
-        let size: Option<String> = params
+        let size: Option<String> = if params.smallest_size_possible {
-            .size_override
+            Some(SMALLEST_IMAGE_SIZE.to_owned())
-            .or_else(|| image_generation_config.size.clone());
+        } else {
+            image_generation_config.size.clone()
+        };
 
         let request = ImagesBody {
             model: Some(image_generation_config.model_id.to_owned()),

src/controller/image/generation.rs

@@ -12,9 +12,6 @@ use crate::strings;
 use crate::utils::mime::get_file_extension;
 use crate::{Bot, entity::MessageContext};
 
-// We may make this configurable (per room, etc.) in the future, but for now it's hardcoded.
-const STICKER_SIZE: &str = "256x256";
-
 pub async fn handle_image(
     bot: &Bot,
     matrix_link: MatrixLink,
@@ -177,7 +174,7 @@ pub async fn handle_sticker(
     );
 
     let params = ImageGenerationParams::default()
-        .with_size_override(Some(STICKER_SIZE.to_owned()))
+        .with_smallest_size_possible(true)
         .with_cheaper_model_switching_allowed(true)
         .with_cheaper_quality_switching_allowed(true);
 

src/strings/usage.rs

@@ -64,7 +64,7 @@ To create a sticker, send a command like `%command_prefix% sticker A huge bowl o
 
 The difference from **creating images** is that the bot will:
 
-- create a smaller-resolution image (`256x256`) - smaller/quicker, but still good enough for a sticker
+- create a smaller-resolution image (as small as the model allows) - smaller/quicker, but still good enough for a sticker
 - potentially switch to a different (cheaper or otherwise more suitable) model, if available
 - post the image directly to the room (as a reply to your message), without starting a threaded conversation
 "#;