feat: Add image support for Claude, OpenAI and Google (#268)

* feat: Add image support for Claude
* feat: Add images support for OpenAI
* feat: Add support images for google ai
* refactor: Separate ImageComponent
* feat: Add attach image button
* feat: Add support image for Mistral provider
* feat: Add support images for OpenAI compatible providers
* feat: Add support images for Ollama
This commit is contained in:
Petr Mironychev
2025-11-20 15:49:39 +01:00
committed by GitHub
parent ce9e2717d6
commit 55b6080273
41 changed files with 860 additions and 93 deletions

View File

@ -41,34 +41,55 @@ public:
if (context.history) {
for (const auto &msg : context.history.value()) {
if (msg.role != "system") {
// Handle thinking blocks with structured content
if (msg.isThinking) {
// Create content array with thinking block
QJsonArray content;
QJsonObject thinkingBlock;
thinkingBlock["type"] = msg.isRedacted ? "redacted_thinking" : "thinking";
// Extract actual thinking text (remove display signature)
QString thinkingText = msg.content;
int signaturePos = thinkingText.indexOf("\n[Signature: ");
if (signaturePos != -1) {
thinkingText = thinkingText.left(signaturePos);
}
if (!msg.isRedacted) {
thinkingBlock["thinking"] = thinkingText;
}
if (!msg.signature.isEmpty()) {
thinkingBlock["signature"] = msg.signature;
}
content.append(thinkingBlock);
messages.append(QJsonObject{{"role", "assistant"}, {"content", content}});
} else {
// Normal message
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
if (msg.role == "system") continue;
if (msg.isThinking) {
QJsonArray content;
QJsonObject thinkingBlock;
thinkingBlock["type"] = msg.isRedacted ? "redacted_thinking" : "thinking";
QString thinkingText = msg.content;
int signaturePos = thinkingText.indexOf("\n[Signature: ");
if (signaturePos != -1) {
thinkingText = thinkingText.left(signaturePos);
}
if (!msg.isRedacted) {
thinkingBlock["thinking"] = thinkingText;
}
if (!msg.signature.isEmpty()) {
thinkingBlock["signature"] = msg.signature;
}
content.append(thinkingBlock);
messages.append(QJsonObject{{"role", "assistant"}, {"content", content}});
} else if (msg.images && !msg.images->isEmpty()) {
QJsonArray content;
if (!msg.content.isEmpty()) {
content.append(QJsonObject{{"type", "text"}, {"text", msg.content}});
}
for (const auto &image : msg.images.value()) {
QJsonObject imageBlock;
imageBlock["type"] = "image";
QJsonObject source;
if (image.isUrl) {
source["type"] = "url";
source["url"] = image.data;
} else {
source["type"] = "base64";
source["media_type"] = image.mediaType;
source["data"] = image.data;
}
imageBlock["source"] = source;
content.append(imageBlock);
}
messages.append(QJsonObject{{"role", msg.role}, {"content", content}});
} else {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
}
}
}

View File

@ -46,7 +46,29 @@ public:
QJsonObject content;
QJsonArray parts;
parts.append(QJsonObject{{"text", msg.content}});
if (!msg.content.isEmpty()) {
parts.append(QJsonObject{{"text", msg.content}});
}
if (msg.images && !msg.images->isEmpty()) {
for (const auto &image : msg.images.value()) {
QJsonObject imagePart;
if (image.isUrl) {
QJsonObject fileData;
fileData["mime_type"] = image.mediaType;
fileData["file_uri"] = image.data;
imagePart["file_data"] = fileData;
} else {
QJsonObject inlineData;
inlineData["mime_type"] = image.mediaType;
inlineData["data"] = image.data;
imagePart["inline_data"] = inlineData;
}
parts.append(imagePart);
}
}
QString role = msg.role;
if (role == "assistant") {

View File

@ -74,7 +74,31 @@ public:
if (context.history) {
for (const auto &msg : context.history.value()) {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
if (msg.images && !msg.images->isEmpty()) {
QJsonArray content;
if (!msg.content.isEmpty()) {
content.append(QJsonObject{{"type", "text"}, {"text", msg.content}});
}
for (const auto &image : msg.images.value()) {
QJsonObject imageBlock;
imageBlock["type"] = "image_url";
QJsonObject imageUrl;
if (image.isUrl) {
imageUrl["url"] = image.data;
} else {
imageUrl["url"] = QString("data:%1;base64,%2").arg(image.mediaType, image.data);
}
imageBlock["image_url"] = imageUrl;
content.append(imageBlock);
}
messages.append(QJsonObject{{"role", msg.role}, {"content", content}});
} else {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
}
}
}
@ -90,7 +114,7 @@ public:
" {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
" ]\n"
"}\n\n"
"Supports system messages and conversation history.";
"Supports system messages, conversation history, and images.";
}
bool isSupportProvider(LLMCore::ProviderID id) const override
{

View File

@ -76,7 +76,19 @@ public:
if (context.history) {
for (const auto &msg : context.history.value()) {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
QJsonObject messageObj;
messageObj["role"] = msg.role;
messageObj["content"] = msg.content;
if (msg.images && !msg.images->isEmpty()) {
QJsonArray images;
for (const auto &image : msg.images.value()) {
images.append(image.data);
}
messageObj["images"] = images;
}
messages.append(messageObj);
}
}
@ -88,11 +100,12 @@ public:
"{\n"
" \"messages\": [\n"
" {\"role\": \"system\", \"content\": \"<system prompt>\"},\n"
" {\"role\": \"user\", \"content\": \"<user message>\"},\n"
" {\"role\": \"user\", \"content\": \"<user message>\", \"images\": [\"<base64>\"]},\n"
" {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
" ]\n"
"}\n\n"
"Recommended for Ollama models with chat capability.";
"Recommended for Ollama models with chat capability.\n"
"Supports images for multimodal models (e.g., llava).";
}
bool isSupportProvider(LLMCore::ProviderID id) const override
{

View File

@ -42,7 +42,31 @@ public:
if (context.history) {
for (const auto &msg : context.history.value()) {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
if (msg.images && !msg.images->isEmpty()) {
QJsonArray content;
if (!msg.content.isEmpty()) {
content.append(QJsonObject{{"type", "text"}, {"text", msg.content}});
}
for (const auto &image : msg.images.value()) {
QJsonObject imageBlock;
imageBlock["type"] = "image_url";
QJsonObject imageUrl;
if (image.isUrl) {
imageUrl["url"] = image.data;
} else {
imageUrl["url"] = QString("data:%1;base64,%2").arg(image.mediaType, image.data);
}
imageBlock["image_url"] = imageUrl;
content.append(imageBlock);
}
messages.append(QJsonObject{{"role", msg.role}, {"content", content}});
} else {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
}
}
}

View File

@ -42,7 +42,31 @@ public:
if (context.history) {
for (const auto &msg : context.history.value()) {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
if (msg.images && !msg.images->isEmpty()) {
QJsonArray content;
if (!msg.content.isEmpty()) {
content.append(QJsonObject{{"type", "text"}, {"text", msg.content}});
}
for (const auto &image : msg.images.value()) {
QJsonObject imageBlock;
imageBlock["type"] = "image_url";
QJsonObject imageUrl;
if (image.isUrl) {
imageUrl["url"] = image.data;
} else {
imageUrl["url"] = QString("data:%1;base64,%2").arg(image.mediaType, image.data);
}
imageBlock["image_url"] = imageUrl;
content.append(imageBlock);
}
messages.append(QJsonObject{{"role", msg.role}, {"content", content}});
} else {
messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
}
}
}
@ -58,7 +82,8 @@ public:
" {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
" ]\n"
"}\n\n"
"Works with any service implementing the OpenAI Chat API specification.";
"Works with any service implementing the OpenAI Chat API specification.\n"
"Supports images.";
}
bool isSupportProvider(LLMCore::ProviderID id) const override
{